ATM Connection Admission Control using Modular Neural Networks

Neural networks, such as multi-layer perceptron (MLP) networks which converge slowly, have been applied for tra c and congestion control in ATM networks. In this paper, we present a Connection Admission Control (CAC) scheme using modular and hierarchical neural networks for predicting the resulting cell loss rate (CLR) when calls are accepted. The fast learning and accurate predictions obtained using this architecture is shown to produce near zero CLR while maintaining a high throughput.