Autonomous QoS Assurance in 5G RAN Slicing using Hierarchical Deep Reinforcement Learning

5G Radio Access Network (RAN) slicing enables the creation of multiple, logically separate networks on a single shared physical infrastructure. This supports services with widely varying Quality of Service (QoS) needs. However, the dynamic and complex nature of 5G makes autonomous resource allocation across network layers very challenging. Our research offers new solutions to the problem of providing quality services to multiple users at the same time through the proposed HDRL (Hierarchical Deep Reinforcement Learning) technique. It breaks down the complex resource allocation issue into two parts: a strategic division managed by the high-tier controller to decide on resource budgets and inter-slice allocation, and low-tier controllers for tactical intra-slice management. This distributed approach enhances the overall quality of service for users in any slice, especially for the Ultra Low Latency Slice, which depends on high communication speeds. This is achieved without compromising the speeds provided to the enhanced Mobile Broadband slices. This innovative approach is the first of its kind. It demonstrates the network’s ability to operate more efficiently without human intervention than traditional methods, which rely on complex monolithic network and resource management strategies.