Deep Learning for Network Slicing and Self-Healing in 5G Systems

The 5th generation cellular network works on a user-centric methodology rather than operator-centric as in 3G or service-centric as seen for 4G. Efficient allocation of resources is possible in 5G networks, something which was not feasible in the previous network generations. The network was equally allocated to all users, whereas it should be allocated depending on the usage. A user playing AR/VR games should be given more bandwidth than a user who is just sending text messages. The main objective of this paper is to develop a solution that can manage the network slice for incoming network requests from unidentified devices. We have compared various machine learning algorithms based on their accuracy to predict the network slice. Also, a Hybrid CNN-LSTM Deep Learning model is proposed for understanding user usage patterns and time series based forecasting of slice utilization, active users, resource usage and workload. The Concept of Self-Healing Networks for better Quality of Experience (QoE) and fault detection, anomaly detection diagnosis is discussed.