A Neural Network Approach to Effective Bandwidth Characterization in ATM Networks

Many call admission control schemes for ATM-type networks focus on the cell loss rate as the exclusive QoS metric and therefore base their Eb (EEective bandwidth) schemes on cell-loss rate approximations. We use simulation data to train an adaptive logic network (ALN) to estimate cell loss and delay; these estimates can then be used to compute eeective bandwidths to satisfy both cell loss and delay. Results indicate that the ALN model is simple, computationally eecient, and suuciently accurate for practical use.