An optimal neural network based call admission control protocol for high-speed networks: modeling and analysis

Presents the development, modeling, analysis and performance evaluation of an optimal neural network based call admission control (CAC) protocol for high-speed Broadband Integrated Services Digital Networks. Thus neural network based CAC protocol simultaneously satisfies two quality of service (QoS) parameters, namely cell loss ratio (CLR) and delay. The protocol presented is suitable for "on-line" application and provides efficient network resource (buffer space and link capacity) utilization. The end-to-end delay and CLR values are divided among the nodes of the connection. For Poisson and ON-OFF sources, M/D/1/K and MMPP/D/1/K queuing models are developed. Analytical expressions at nodes for CLR and delay have are developed based on these models. These analytical expressions are used to develop a CAC protocols. The principles behind the Kohonen neural networks were used in the development and software implementation of this CAC protocol.