Deep Learning based User Slice Allocation in 5G Radio Access Networks

Abstract

Network slicing is proposed as a new paradigm to serve the plethora of 5G services on a shared infrastructure. Within this context, a Radio Access Network (RAN) slice is considered as the proportion of physical spectrum resources to be served to third parties. Interestingly, 3GPP standardized options of RAN processing dis-aggregation into network functions while enabling their placement whether in distributed or centralized locations. The adoption of an end-to-end RAN slicing raises new challenges related to the allocation efficiency of joint radio, link and computational resources. To deal with the stringent latency requirements of 5G services, we propose, in this paper, a Deep Learning based approach for User-centric end-to-end RAN Slice Allocation scheme. It can decide in real-time, to jointly allocate the amount of radio resources and functional split for each end- user. Our proposal satisfies end-user's requirements in terms of throughput and latency, while minimizing the infrastructure deployment cost.