A novel neural estimator for call admission control and buffer design in ATM network

In call admission control of ATM network, it is difficult for the conventional method to judge the accepting boundary accurately and dynamically, owing to the imprecise description of the traffic parameters and the different requirement of the allowed QoS. In this paper we propose a novel neural network structure as an intelligent control scheme to perform ATM admission control. The neural estimator can learn the probability distribution of the CLR and thus can control the ATM traffic very accurately and dynamically. The disperse structure of the neural estimator makes it easy to learn and operate. The trained neural network can also be used as a buffer estimator in the reference design of ATM system. The simulation results show the advantage of this neural estimator.