A hierarchical structure neural network for aggregate bandwidth allocation of heterogeneous sources in ATM networks

This paper proposes an application of neural network for aggregate bandwidth allocation of heterogeneous sources in ATM networks. A feedforward hierarchical neural network and backpropagation training method are used for recognizing the relation between trained input and bandwidth output calculated from exact analysis. The results show that the proposed neural network can allocate more accurate bandwidth than the approximate method-asymptotic analysis with a reasonably fast speed compared to that of the approximation method