A Hierarchical Neural Optimization Structure for Access Control and Resource Sharing in Integrated Communication Networks

Abstract The access to a multiservice broadband synchronous time division multiplexer is considered, where hybrid frames are used to carry two basic traffic types: a circuit-switched isochronous and a packet-switched asynchronous one, generated by several users. Each user has an assigned portion of the total available bandwidth, in terms of slots/frame, which is dynamically allocated between the two traffic types at the user premise, by means of a local randomized decision rule. Its structure is determined by a neural network of the nonlinear perceptron type. A two-level optimization problem is defined, through which the users independently adjust their local parameter values (i.e., the synaptic weights of the neural networks) and a central controller reassigns the bandwidth partitions, that play the role of coordination variables. At the end some simulation results and conclusions on the usefulness of nonstationary and stationary neural networks to feedback control of ISDN are provided.