A neural based call admission control in ATM networks to provide multiple QOS

In high-speed asynchronous transfer mode (ATM) networks several classes of traffic streams with widely varying traffic characteristics and quality of service requirements are statistically multiplexed and share common switching and transmission resources. In such an environment we need a traffic control scheme which manages the required QOS of each class individually. We propose a neural network (NN) approach to estimate the cell loss probability (CLP) of bursty sources for call admission control (CAC) purpose in ATM environment. Training data set are obtained from a CLP method advocated by Miyao (see IEEE GLOBECOM´93, p.1398-1403, 1993), and this training is done off-line to estimate the CLP in a real time environment. We choose this method as it considers the CLP due to long-term fluctuations in the aggregate cell rate as well as short-term fluctuations in cell interarrival time and hence provides a higher accuracy of loss probabilities than traditional methods. The accuracy of the training set is crucial in the NN based training scheme to design a robust call admission controller. We find that our scheme performs consistently well in withstanding the changes in the burst duration parameters and is also suitable from the point of view of individual call quality. In order to discuss performance aspects the scheme has been compared with other cell loss estimation methods. Our simulation shows that the NN approach outperforms the conventional methods in terms of accuracy