Fuzzy logic for controlling call admission in ATM networks

An essential function of ATM is Connection Admission Control (CAC). To achieve higher network utilisation, bursty calls are statistically mul-tiplexed on a common link. We present a new CAC scheme based on Fuzzy Logic (FL) and Artiicial Neural Networks (ANNs) to exploit this key feature of ATM networks. Deening a hybrid intelligent system enables us to take advantage of both the learning capabilities of ANNs and the interpretability properties of FL. The ANN is used in the learning phase to automatically tune the fuzzy system (i.e. to de-ne the fuzzy rules and the membership functions) whilst during the control phase, the fuzzy system forecasts the QoS values. The CAC procedure is able to determine on line the Cell Loss Probability (CLP) that a connection will exhibit when accepted into the ATM network. Because of its possible hardware implementation as well as its adaptive and learning capabilities, this scheme constitutes a good candidate for a robust real-time CAC. Submission area: AI techniques for network and services management.