SCAN: ML-based Slice Congestion and Admission Network Controller

Network slicing enables 5G/6G networks to support Ultra-Reliable Low-Latency Communication (ULLC), enhanced Mobile Broadband (eMBB) and Massive Machine-Type Communication (mMTC). However, while this virtual networking technology enhances network efficiency, it also adds substantial signaling overhead. Maintaining sub-millisecond latency and managing dense deployments require continuous signaling at high resolution, which keeps hardware components active, leading to increased energy consumption. In this paper, we introduce a novel network controller that manages slice congestion and admission, designed to meet flexible Quality-of-Experience requirements for both priority and non-priority traffic. Utilizing metadata from Internet of Things (IoT) device applications and network characteristics, we introduce adaptability and elasticity features, enabled by transfer and reinforcement learning, significantly lowering signaling overhead and network resources. Further, analytical results show the proposed framework effectively reduces rejection rates and congestions across varying mMTC and eMBB traffic loads.