A Predictive Intelligent Traffic Controller for Broadband Multimedia IP Networks

Abstract

The paper proposes a predictive intelligent traffic controller for broadband multimedia IP networks with quality-of-service (QoS) provisioning. The predictive intelligent traffic controller considers a user's requested traffic characteristics, QoS requirements, and predicted system performance measures to determine its connection admission. A pipelined recurrent neural network with extended recursive least square learning algorithm (PRNN/ERLS) is employed to predict the system congestion status and the packet loss probability of the broadband multimedia systems. The PRNN/ERLS possesses an infinite memory of previous signals and appropriately weights them to precisely capture the correlation among signals of these performance measures; and thus the effect on the congestion status and packet loss probability can be well captured and even estimated in advance. Simulation results show that the proposed predictive intelligent traffic controller can achieve higher system utilization than the conventional non-predictive intelligent traffic controller by about 5.6% at light-to-heavy traffic transient periods.