Dynamic Resource Allocation for Radio Access Network Slicing using CNN-A2C Algorithm

The conventional generic network architecture is no longer efficient for addressing the diverse and evolving service requirements of 5G technologies. Radio Access Network (RAN) Slicing involves partitioning the resources and functionalities of the RAN for the provisioning of various services with specific requirements. Dynamic Resource Allocation is a crucial component of RAN Slicing. Previous studies have shown that Deep Reinforcement learning (DRL) algorithms exhibit promising results by providing efficient resource allocation schemes. However, a prevalent limitation in existing research is the insufficient consideration of a wide range of features within the state space. To address this, we leverage the feature extraction capabilities of convolutional networks (CNN) together with the DRL advantage actor-critic (A2C) and propose the CNN-A2C algorithm. CNN-A2C is used to allocate bandwidth to network slices ensuring the Quality of Experience (QoE) and optimizing the Spectral Efficiency (SE). Simulation results demonstrate that the CNN-A2C algorithm significantly improves the SE while effectively meeting the QoE requirements.