Hierarchical neuro-fuzzy call admission controller for ATM networks

Neuro-fuzzy systems, have recently been applied to the problem of Call Admission Control (CAC) in ATM networks. Their advantage as opposed to classical schemes, based on the modeling of sources connected to the network, is their capability to learn the behavior of data sources through training process, and to predict the future behavior based on the experience acquired in the learning process. They also allow to extract the knowledge acquired through the learning process, and to integrate some initial knowledge about the system behavior in order to improve the performance and training speed. In this work, a hierarchical neuro-fuzzy CAC controller for ATM network is proposed, based on GARIC architecture, introduced by Berenji and Khedkar. It contains a neural network as a critic, using the reinforcement learning scheme, and three fuzzy sub-networks, controlling cell loss ratio, queue size and link utilization in the ATM multiplexer, respectively. The final CAC decision is obtained as a weighted combination of the decisions generated by the fuzzy sub-networks. The implementation of the controller is performed, and the results are analyzed in terms of network utilization, and the generated knowledge base. 1 The work was performed when the author was with EE Dept., %R÷D]LoL 8QLYHUVLW\ ,VWDQEXO 7XUNH\