Dynamic Link Classification Based on Neuronal Networks for QoS Enabled Access to Limited Resources

Abstract

In this paper, the authors present a study of a network fingerprinting classification using monotone multilayer perceptron neuronal networks. It is part of an overall performance engineering approach. The classification is used to increase the performance of an active queue management on the quality of service for a next generation public safety communication system based on an IP overlay network. This network combines heterogeneous communication networks and technologies to increase the overall systems performance. Public safety users have higher requirements regarding coverage, data rates and quality of service than standard commercial ones. Main challenge for this study is the optimization of the overall system for voice group communication, which is still the most important communication within public safety scenarios. This paper shows that with the given parametrization, an ensemble of multi-layer perceptrons gives a satisfactory classification probability, if a setup of three technologies (EDGE, UMTS and LTE) is assumed to be in usage as communication technologies. This setup is practicable enough to have a chance to be implemented in a future system.