An End-to-End Network Slicing Algorithm Based on Deep Q-Learning for 5G Network

Abstract

As one of key technologies of the fifth-generation (5G) communication system, network slicing can share the underlying infrastructure with different application requirements and ensure that the slices can be isolated from each other. This paper proposes an end-to-end (E2E) network slicing resource allocation algorithm based on Deep Q-Networks (DQN), which is suitable for multi-slice and multi-service scenarios. This algorithm jointly considers the radio access network slices and core network slices to dynamically allocate resources to maximize the number of access users. First we build such a model, which is a mixed integer programming problem and it needs to be dynamically adjusted according to the changes of environment. We propose to use DQN algorithm to solve this problem, which can perceive changes in the environment and make dynamic decisions. Under each decision, we need to calculate the reward value of DQN, so we divide the problem into the core side and the access side. Then the dynamic knapsack algorithm and the link mapping algorithm are used to obtain the reward. The simulation results show that the average access rate of DQN scheme is higher than 97%. Compared with the optimal allocation scheme of access side, the average access rate is increased by 9% for delay constrained slices and 5% for rate constrained slices in a dynamic environment.

Keywords

• 5G network
• network slicing
• resource allocation
• deep Q-networks