New training pattern selection method for ATM call admission neural control

A new training pattern selection method for adaptive neural control using backpropagation learning is presented. When applying this method to an asynchronous transfer mode (ATM) call admission control, some advantages are observed: a very small training pattern table is sufficient, the learning is independent of observed data, and the controller is easily adaptable to traffic changes. The proposed neural control model is analysed by computer simulations in heterogeneous traffic environments and the results show its effectiveness compared with other methods