LSTM based Dynamic Bandwidth Allocation in EPONs for Future Multimedia Applications

The bandwidth requirement of video type applications such as Internet-enabled HD television and live streaming generated as much traffic as an entire home today, on average, during the past year, because of the realization of the 4K or more video streaming application. Ethernet Passive Optical Networks (EPONs) have emerged as a promising alternative for next-generation broadband access networks that can handle bandwidth-intensive video applications. One of the most primary aspects in successfully allocating bandwidth to ensure Quality-of-Service (QoS) is the Dynamic Bandwidth Allocation (DBA) Scheme. In this paper, we present a Recurrent-Neural-Network (RNN) based on Long-Short-TermMemory (LSTM) to increase bandwidth usage by improving the efficiency of the DBA scheme. Simulation result shows that our proposed DBA scheme is successfully predict the AF GRANT message for the next cycle with the Mean Absolute Error of 0.0321.