CALL ADMISSION CONTROL IN ATM NETWORKS NEUROCOMPUTING APPROACH

Abstract

The traffic control in ATM based B-ISDN has been the subject of vigorous research over the past several years. The design of a suitable ATM traffic controller is considered as a fundamental challenge for the success of B-ISDN. In the present work a neurocomputing Call Admission Control (CAC) algorithm for ATM networks has been simulated. The algorithm presented employs Neural Networks (NNs) to calculate the bandwidth required to support multimedia traffic with multiple QoS requirements. The NN based CAC calculates bandwidth required per call using measurements of the traffic via its count-process, instead of relying on simple parameters such as the peak, average bit rate and burst length. Furthermore, to enhance the statistical multiplexing gain, the controller calculates the gain obtained from multiplexing multiple streams of traffic supported on separate virtual paths (i.e., class multiplexing). In order to simplify the design and obtain small reaction time, the controller is simulated using a hierarchical structure of a bank of small size, parallel NN units. Each unit is a feed forward backpropagation NN that has been trained to learn the complex non linear function relating different traffic patterns, . with the corresponding received. capacity. The neurocomputing approach is effective in achieving more accurate results. This is primarily due to the unique learning and adaptive capabilities of NNs that enable them to extract and memorize rules from previous experience.