On-line learning and dynamic capacity allocation in the traffic management of wireless ATM networks

We consider digital wireless multimedia ATM LANs. A traffic monitoring algorithm (TMA) is utilized to allocate channel capacities dynamically to heterogeneous traffic with time-varying characteristics. The threshold parameters that characterize the operational characteristics of the TMA control the overall system performance, including traffic rejection rates and wasted capacity rates. In this paper an on-line learning algorithm is introduced, to dynamically adapt the threshold parameters of the TMA for attaining a prespecified upper bound on the traffic rejection rate, at minimal cost in wasted capacity rate. The learning algorithm is analyzed and evaluated to expose its superior convergence characteristics