Actor Critic Based Reinforcement Learning for Joint Resource Allocation and Throughput Maximization in 5G RAN Slicing

Abstract

With the advent of fifth generation (5G) mobile communication network slicing technology, the range of application scenarios is expanding significantly. For 5G to function well, it necessitates little delay, a fast rate of data transfer, and the ability to handle a large number of connections. This demanding service requires the allocation of resources in a dynamic manner, while maintaining a very high level of reliability in terms of Quality of Service (QoS).The applications like autonomous driving, telesurgery, etc. have stringent QoS demands and the present design of slices is not suitable for these services. Therefore, latency has been regarded as a crucial factor in the design of the slices. Conventional optimization algorithms often lack robustness and adaptability to dynamic environments, getting stuck in local optima and failing to generalize to varying conditions. Our solution utilizes Reinforcement Learning (RL) to allocate resources to the slices. The utilization of restricted resources can be optimized through the reconfiguration of slices. The ability of RL to acquire knowledge from the surroundings enables our solution to adjust to varying network conditions, enhance the allocation of resources and improve quality of service over a period of time for different network slices. This study introduces the Deep Actor Critic Reinforcement Learning- Network Slicing (DACRL-NS) technique, which utilizes Deep Actor Critic Reinforcement learning for efficient resource allocation to network slices. The objective is to achieve optimal throughput in the network. If the slices fail to meet the minimum criteria, they will be omitted from the allocation. With increasing training episodes, our Actor-Critic algorithm enhances average cumulative rewards and resource allocation efficiency, demonstrating continuous learning and improved decision-making.The simulated suggested system demonstrates an average throughput improvement of 8.92% and 16.36% with respect to the rate requirement and latency requirement, respectively. The data also demonstrate a 17.14% increase in the overall network throughput.