A self‐learning control of atm service quality using neural network

Abstract

AbstractStudies of ATM network are conducted aiming at the integrated processing of multimedia. One of the important problems is to find a control method for the ATM network which can meet various types of requirements for the service quality arising from the diversity of sources connected to the network. For such a problem, the traditional approach by exhaustive analysis and simulation that has been used in the circuit switching network has a problem in that a large number of parameters to be considered in the ATM network complicates the control and it is difficult to deal flexibly with the diversity of sources and the change of traffic characteristics due to the new service.The leaky pattern table method is proposed which can realize a highly accurate control by learning from the data with small generation rate. As an example of application of the proposed method, the call admission control in the ATM node is considered. The detailed construction of the control system is discussed, and the proposed system is evaluated by computer simulation using a basic model.From such a viewpoint, this paper applies the neural network which can realize a high‐speed learning of the multiinput/multi‐output nonlinear function by back propagation to obtain the relation between the load in the ATM network and the communication quality by training. This network is used also to effect the learning control to deal flexibly with the change of the environment.