Dynamic ATM connection admission control based on instantaneous node state using an artificial neural network

It is widely accepted that the future broadband communication network will be based on Asynchronous Transfer Mode (ATM). While ATM allows advantage of statistical multiplexing gain obtained by interleaving multiple asynchronous traffics, due to its high speed and asynchronous transmission, a fast, robust and intelligent control over widely fluctuating traffic is necessary to maintain quality of service goal.; This thesis proposes a dynamic ATM connection admission control (CAC) method using artificial neural network (ANN) which relies not on classification of each call but on instantaneous node state vector and user provided traffic descriptor.; Above technique was implemented and tested using ATM simulator, which simulates real ATM user network interface (UNI) protocol.; Extensive simulations were performed to confirm the effectiveness of the method. It was verified that at connection setup phase decision on connection acceptance can be made not based on the class of the source end system (SES) but on the traffic descriptor of the connection itself. Also, the result showed that performance of switch can be conveniently regulated by adjusting one parameter, called CAC threshold value, which determines the level of traffic influx into the network.