Deep Learning-based Application Specific RAN Slicing for Mobile Networks

Effectively identifying application is desirable for network operators to improve spectrum efficiency and user experience in future mobile networks that are expected to support multiple kinds of applications with different quality of service (QoS) requirements. In this paper, we present a Radio Access Network (RAN) slicing architecture utilizing in-network deep learning to apply application specific radio spectrum scheduling. We use a small number of customized supervising phones to generate training data in real-time and apply deep learning at the packet gateway (P-GW), where we tag the downlink packets with the identified application name and transmit them to eNB for application specific spectrum scheduling. The preliminary experimental results show the feasibility and the efficiency of the proposed application specific RAN slicing.