Optimizing communication and computational resource allocations in network slicing using twin-GAN-Based DRL for 5G hybrid C-RAN

Abstract

The network slicing defined from 3GPP Rel. 15 is one important feature and function for 5G networks. In this paper, a new machine learning scheme is proposed by extending existing generative adversarial network (GAN) based deep reinforcement learning (DRL) result, namely Twin-GAN-based DRL (TGDRL) scheme, by utilizing two GAN-based DRLs to jointly allocate wireless bandwidth resources and computational resources. Existing resource allocation results are just only consider the bandwidth allocation, or just only consider the computational resource allocation. The main contribution of the proposed TGDRL scheme is to simultaneously investigate the bandwidth allocation and computational resource allocation by utilizing a multi-objective optimization algorithm, which aims to improve the efficiency of spectrum and reduce the consumption of computational resources. In our simulation, the total delay, the spectral efficiency, and the computational consumption of our proposed scheme is improved by 10.2%, 15.7%, and 12.8%, compared to existing schemes.