Neural networks for admission control in an ATM network
Ernst Nordström, Olle Gällmo, Lars Asplund, Mats Gustafsson, Bo Eriksson · 1994
This paper presents an artificial neural network (ANN) approach to link admission control in ATM communication networks. Three different ANN models for implementation of link quality of service formulas, based on a heterogeneous fluid--flow queueing model, are presented. The first model uses predefined peak rate parameters, the second model is based on a state interpretation of aggregated link traffic, and the third model employs a form of statistical pre-processing of the traffic parameters.It is argued that the ANN must implement a function which is invariant to every permutation of the traffic descriptor arguments. This constraint is met by the third ANN model and the experiments presented also suggests that the pre-processing performed is benificial for generalization situations.