Efficient call-connection model for multipoint connections in an ATM/B-ISDN network using fuzzy neural networks

This concept paper describes a methodology to increase the efficiency and throughput of a multipoint call connection system in an ATM/B-ISDN network using fuzzy-neural network theory. Analysis of the B-ISDN network reveals that it is possible to incorporate a simplified fuzzy neural scheme into the call-connection model that can effect an efficient system. The network consists of interior nodes that interface to other nodes and exterior nodes that interface to the clients and network clients which are all interconnected by fibre optic links. Each node in the network contains an ATM packet switch now modified to include the fuzzy-neural scheme. This is expected to increase the efficiency of the switch architecture. This architecture now consists of the fuzzy-neural broadcast packet network (FNBPN). This FNBPN consists of a modified fuzzy-neural copy network (FNCN) interconnected in a maze to a fuzzy-neural routing network (FNRN). The arbitrary nature of the path of the point-to-point cells is controlled more systematically by the control processor (CP). This CP through a learning process configures the switch hardware to route incoming cells to the corresponding outgoing-cells and effecting efficient connections. The neural net learning function uses the backpropagation to learn the needed membership functions from a set of training examples if the output type is continuous and the learning method is supervised and the probabilistic method if the output type is discrete and the learning method is supervised; otherwise the popular Kohonen/LVQ is used as an unsupervised learning method. It can be concluded that the use of fuzzy-neural networks is beneficial in complex ATM processing.