LSTM-based QoS prediction for 5G-enabled Connected and Automated Mobility applications

Determining whether the network can provide the required resources -and as a result the required Quality of Service (QoS)- for Connected and Automated Mobility (CAM) applications is crucial, as human safety is involved. In order to identify potential QoS deterioration, Machine Learning, which is currently considered a momentous enabler for the 5G and beyond systems, can be exploited in order to timely predict such changes in the QoS and proactively notify the CAM application in order for it to flexibly adapt. This paper explores the validity and viability of Recurrent Neural Networks (RNNs) -and more specifically LSTMs- as potential enablers for realizing such predictions. To this end, a novel QoS prediction mechanism is presented that uses an LSTM architecture and network-related QoS metrics, in order to identify specific patterns and predict in advance the QoS that will be available in the near future to a Connected and Automated Vehicle (CAV). Simulation-based evaluation of the proposed scheme prove its potential gains for the CAM domain.