Growing and pruning (GAP) RBF networks for call admission control in ATM traffic management

Abstract

This thesis presents a study on the use of recently developed neural networks MRAN (Minimal Resource Allocation Network) and GAP (Growing and Pruning neural network) for the performance enhancement of Call Admission Control in Asynchronous Transfer Mode (ATM) networks. GAP and MRAN generate a minimal radial basis function neural network by adding and pruning hidden neurons based on input data and are ideal for online adaptive control of fast time-varying non-linear systems. The use of GAP and MRAN in the study of call admission control schemes is new. The fast learning and accurate predictions obtained with the neural networks are shown to make better call admission control decisions under heavy traffic situations compared to conventional schemes.