A Deep-Learning-Based Radio Resource Assignment Technique for 5G Ultra Dense Networks

Recently, deep learning has emerged as a state-of-the-art machine learning technique with promising potential to drive significant breakthroughs in a wide range of research areas. The application of deep learning for network traffic control, however, remains immature due to the difficulty in uniquely characterizing the network traffic features as an appropriate input and output dataset to the learning structures. The network traffic features are anticipated to be even more dynamic and complex in the UDNs of the emerging 5G networks with high traffic demands coupled with beamforming and massive MIMO technologies. Therefore, it is critical for 5G network operators to carry out radio resource control in an efficient manner instead of adopting the simple conventional F/TDD. This is because the conventional uplink-downlink configuration change in the existing dynamic TDD method, typically used for resource assignment in beamforming and massive-MIMO-based UDNs, is prone to repeated congestion. In this article, we address this issue and discuss how to leverage the deep LSTM learning technique to make localized prediction of the traffic load at the UDN base station (i.e., the eNB). Based on localized prediction, our proposed algorithm executes the appropriate action policy a priori to avoid/alleviate the congestion in an intelligent fashion. Simulation results demonstrate that our proposal outperforms the conventional method in terms of packet loss rate, throughput, and MOS.