Channel assignment for cellular radio using neural networks

Abstract

The channel assignment problem, i.e. the task of assigning channels to radio cells in a spectrum-efficient way, is solved by a neural network algorithm. This algorithm is inherently parallel and does not rely on a particular structure of the interference graph. The results obtained so far indicate that the algorithm can be used to obtain an optimum solution. It was applied successfully for inhomogeneous interference conditions and channel demand. Cochannel and cosite constraints were taken into account, and the extension to any other technical restrictions will be possible in an obvious way. The examples studied to date are of a relatively small size; the question remains of how the algorithm behaves when applied to larger and more complex examples. The disadvantages of the algorithm are its long calculation time compared to graph coloring algorithms and the difficulty of finding appropriate parameters. However, the algorithm was not optimized for speed, and the parameter search may be a question of experience.<>