Performance analysis of emulated dynamic multi-service UMTS core networks using clustering and neural modelling

In this paper, we present the modelling of the dynamic behaviour of an ATM-based, multi-service UMTS core network with calls that belong to one of four service classes and arrive randomly. Arriving calls are granted service based on specific service class, required maximum and minimum bandwidth, and available network resources. Performance of priority-based dynamic capacity allocation, suitable for the wireless system supporting ATM-like traffic is analysed. Scheduling of the ATM cell transmission in each uplink TDMA frame is based on a priority scheme. GoS (blocking probability) and QoS (throughput) parameters for bandwidth sharing policy (BSP) are considered, and partial overlapped transmission link (POL) is implemented. In the modelling, the clustering procedure is developed based on Markov reward models (MRM), enhanced by the self-organizing vector quantification (VQ) and neural modelling. The optimal link occupancy probability distribution is calculated using the neural network, trained with Kohonen rules. Simulation and numerical results are shown. The efficiency of the link occupancy probability density function, applied for the network performance analysis, is confirmed by obtained numerical results. The approach proved to be useful and practical, allowing the simulations to be performed faster, easier, and less costly in resources. Although in this work we analyse performance of priority-based dynamic capacity allocation, suitable for the wireless network carrying ATM traffic, the results can be easily extended to ATM-like traffic, i.e. traffic of a similar nature and with comparable GoS/QoS parameters.