Neural network control of shared ATM buffer

Assuming the shared buffer in ATM node with bursty input traffic, we have derived a new competitive neural network algorithm for buffer management. The algorithm is an extended version of the previous one, based on input prebuffers solution. The cell loss rate obtained is considerably lower comparing to the round-robin control in such a case. The interarrival times of output streams are considered by fractal/multifractal analysis, as well.