Reinforcement Learning-Based Network Slicing Scheme for Optimized UE-QoS in Future Networks

An end-to-end (E2E) network slicing (NS) scheme for heterogeneous network (HetNet) is proposed in which the number of slices and instances of various network functions (NFs) are optimized contingent on the number of users (UEs) and their quality of service (QoS) requirements. The objective of the scheme is to empower future generation networks by considering control signaling in the control plane as well as the UE traffic in the user plane of the core network (CN). We formulate a joinT UE assOciation, wiReless bandwidth allocation, sliCe formation, slice assignment, virtual network function (VNF) placement, computing resource allocation, link assignment, and link bandwidtH allocation (TORCH) problem to minimize the E2E task completion time of all UEs while considering both control signaling and UEs’ traffic. Since TORCH is a mixed-integer nonlinear problem, to tackle the problem, we decompose it into two sub-problems: the link assignment problem and the UE Association, resource allocation, Slice formation, Slice AssIgnment, and VNF pLacement (ASSAIL) problem. The ASSAIL problem comprises both the core network (CN) and radio access network (RAN), and they do not compete for resources, so we decompose it into two sub-problems: the RAN problem and the CN problem. We use Dijkstra’s algorithm and a deep Q-learning network (DQN) based reinforcement learning method to iteratively solve the two sub-problems. Simulation results have confirmed the effectiveness of our proposed scheme in tackling the TORCH problem.