Nonlinear auto-regressive neural network model for forecasting Hi-Def H.265 video traffic over Ethernet Passive Optical Networks

Abstract

Video bandwidth forecasting can help optimize the transmission of video traffic over optical access networks. In this paper, we propose the use of a nonlinear auto-regressive (NAR) neural network model for forecasting H.265 video bandwidth requirements to optimize video transmission within Ethernet Passive Optical Networks (EPONs). The video's constituent I, P, and B frames are forecast separately to improve model forecasting accuracy. The proposed forecasting model is able to forecast H.265 encoded High-Definition videos with an accuracy exceeding 90%. In addition, using the video bandwidth requirement predictions as grant requests within EPONs improved the efficiency of dynamic bandwidth allocation (DBA). The use of nonlinear auto-regressive neural network grant sizing predictions within EPONs reduced the video packet queueing delay significantly when the network was saturated near capacity.