Use of neural networks in policing bursty traffic sources in Broadband ISDN ATM networks

The paper discusses the possibility of using artificial neural networks to effectively police the user traffic irrespective of its nature in Broadband ISDN ATM networks. A multi-layer neural network is used to control the user's traffic parameters within their declared values in order to protect the network's resources. This is achieved by actually employing two neural networks in parallel. One is trained to learn (memorise) the characteristics of a compliant source, whereas the other is used to accurately estimate the characteristics (p.d.f) of the actual offered user traffic. The two are compared and in case of any incompliance an error signal is generated and the user forced to reduce or modify its traffic generation pattern or else risk some of its cells (violating) being discarded. The paper discusses the possible use of neural networks in policing user traffic in Broadband ATM networks.