A tenant-driven slicing enforcement scheme based on Pervasive Intelligence in the Radio Access Network

In Beyond 5G mobile networks, the network slicing paradigm offers the possibility of sharing the network infrastructure among different tenants: the tenant declares communication service requirements and the Infrastructure Provider configures (potentially on-demand) the corresponding network slice instances. The management of network slicing in the core network has been deeply investigated in the current scientific literature. On the contrary, handling network slicing in the Radio Access Network still represents an open and very challenging research topic, mostly due to the unpredictable variability of the wireless channel, network dynamics and heterogeneity, slice isolation, as well as different Quality of Service requirements of various services. In order to achieve an important step forward in this direction, this paper proposes a tenant-driven Radio Access Network slicing enforcement scheme based on Pervasive Intelligence. The proposed approach grounds its roots in the Pay for What You Get paradigm: it promises to avoid the radio resources over-provisioning while saving bandwidth. To achieve these goals, Artificial Intelligence mechanisms are innovatively and pervasively integrated into some key functionalities of both Infrastructure Provider and tenants. On the one hand, the Infrastructure Provider exploits a Deep Learning scheme (i.e., convolutional autoencoder) to compress the information on network resources and connectivity and share the actual (but hidden through compression) network status with the tenants. On the other hand, each tenant implements a Deep Reinforcement Learning algorithm (i.e., Deep Deterministic Policy Gradient) to dynamically adapt bandwidth requests according to its own users’ requirements. The outcomes of this algorithm are then used by the Infrastructure Provider to effectively enforce network slicing at the Radio Access Network level. Computer simulations investigate the proposed approach in a realistic scenario, which jointly embraces enhanced Mobile BroadBand, Advanced Driver Assistant Systems, and best-effort applications. Obtained results demonstrate the effectiveness of the proposal against conventional resource allocation methods