Bitrate management in ATM systems using recurrent neural networks

Abstract

Future wide as well as local area integrated broadband communication networks will be based on the asynchronous transfer mode (ATM). ATM allows the exploitation of a statistical multiplexing gain for connections with variable bitrates. On the other hand, an efficient and robust traffic control is necessary to guarantee quality of service objectives. The paper proposes a fast and flexible method to estimate the bitrate which is required for an arbitrary traffic mix to keep the cell loss probability below an acceptable limit. It is based on discrete-time recurrent neural networks and considers the parameters of each connection individually. The used neural networks are described and the performed studies and their results are presented.