Dynamic handoff policy for RAN slicing by exploiting deep reinforcement learning.

It has been widely acknowledged that network slicing is a key architectural technology to accommodate diversified services for the next generation network (5G). By partitioning the underlying network into multiple dedicated logical networks, 5G can support a variety of extreme business service needs. As network slicing is implemented in radio access networks (RAN), user handoff becomes much more complicated than that in traditional mobile networks. As both physical resource constraints of base stations and logical connection constraints of network slices should be considered in handoff decision, an intelligent handoff policy becomes imperative. In this paper, we model the handoff in RAN slicing as a Markov decision process and resort to deep reinforcement learning to pursue long-term performance improvement in terms of user quality of service and network throughput. The effectiveness of our proposed handoff policy is validated via simulation experiments.