Machine learning based bandwidth prediction in tactile heterogeneous access networks

Abstract

We present the use of machine learning to predict bandwidth demand over heterogeneous optical-wireless networks, and propose a novel predictive bandwidth allocation algorithm, termed MLP-DBA, to specifically meet stringent upstream delay requirements in emerging tactile applications. With a trained artificial neural network at the control server located at the central office, the proposed MLP-DBA classifies the status of each optical network unit based on bandwidth demand prediction and makes flexible bandwidth allocation decisions accordingly. Extensive simulations highlight the effectiveness of MLP-DBA in reducing the upstream delay as well as dropped packets.