LSTM-Based DWBA Prediction for Tactile Applications in Optical Access Network

Abstract

Historically, the optical access network (OAN) plays a crucial role of supporting emerging new services such as 4 k, 8 k multimedia streaming, telesurgery, augmented reality (AR), and virtual reality (VR) applications in the context of Tactile Internet (TI). In order to prevent losing connectivity to the current mobile network and Tactile Internet, the OAN must expand capacity and improve the quality of Services (QoS) mainly for the low latency of 1 ms. The optical network has adopted artificial intelligence (AI) technology, such as deep learning (DL), in order to classify and predict complex data. This trend mainly focuses on bandwidth prediction. The software-defined network (SDN) and cloud technologies provide all the essential capabilities for deploying deep learning to enhance the performance of next-generation ethernet passive optical networks (NG-EPONs). Therefore, in this paper, we propose a deep learning long-short-term-memory model-based predictive dynamic wavelength bandwidth allocation (DWBA) mechanism, termed LSTM-DWBA in NG-EPON. Future bandwidth for the end-user is predicted based on NG-EPON MPCP control messages exchanged between the OLT and ONUs and cycle times. This proposed LSTM-DWBA addresses the uplink control message overhead and QoS bottleneck of such networks. Finally, the extensive simulation results show the packet delay, jitter, packet drop, and utilization.