Research Of QoS Optimization Based On Neural Network Prediction For Ethernet Passive Optical Network

Ethernet passive optical network (EPON) is considered as the most accepted optical access network currently. Among them, the dynamic bandwidth allocation (DBA) is used for scheduling coordination mechanism, which is the key technology of quality of service (QoS) of uplink business, the research of DBA is very of great imprortance for enhancing the performance of EPON system.OLT communications with ONU by five kinds of control frames in multi-point control protocol(MPCP). If frames arriving during waiting time in ONU can be predicted, packet delay will be greatly reduced and system uplink efficiency will be improved. The most common prediction schemes are mainly linear predictors, which do not meet the self-similar long range dependence characteristics of network business flow. This paper introduces network business nonlinear prediction to DBA mechanism, formatting a new scheduling mechanism and algorithm combined of report responding and business forecasting, to optimize EPON system QoS performance under uplink.This thesis firstly reviews Ethernet passive optical network system architecture and communication protocol, then analyzes in detail all kinds of existing dynamic bandwidth allocation scheme from two aspects: scheduling mechanism and authorized length algorithm. In-depth analysis of self-similarity long range dependence features of network business are then presented. After explaining the neural network theory, a simulation using MATLAB is carried out which proves the accuracy of neural network prediction of network flow. Then a new algorithm named neural network prediction based dynamic bandwidth allocation algorithm (NPBA) is introduced, and a simulation experiment is established with OPNET tool. Simulation results indicate that the neural network model can realize nonlinear prediction accuracy, and neural network prediction based dynamic bandwidth allocation algorithm (NPBA) can effectively reduce end-to-end delay, restrain delay jitter, optimize the quality of service without affecting system throughput compared with other general DBA algorithm.