Treffer: Optimizing URLLC Resource Allocation in Open-RAN: A Transformer-based Approach to Flexible Numerology

International audience ; Beyond 5G networks are set to address the continuous surge in data traffic. A key enabler is the Open Radio Access Network (Open-RAN) architecture. Open-RAN aims to virtualize and decouple the network components to facilitate network management, enable AI-based control and optimization, and optimize energy efficiency and operator's costs. One of the services in 5G and beyond is Ultra Reliable Low Latency Communication (URLLC) which requires stringent delay and reliability requirements. Hence, allocating resources to satisfy these Quality of Service constraints is challenging. 5G allows for flexible numerology that controls subcarrier spacing, enabling the use of wider subcarriers to achieve shorter transmission times. This reduction in transmission delay must be paired with selecting a suitable Modulation and Coding Scheme (MCS) to meet the reliability requirements. To address this, we formulate a Mixed-Integer Linear Programming (MILP) problem to dynamically allocates Resource Blocks (RBs), MCS, numerology, time slots, and CPU resources per packet to meet URLLC traffic requirements. This dynamic model is evaluated against fixed numerology schemes. It is also compared with semi-dynamic schemes, where the numerology for all packets is unified but changes dynamically every 1ms. Additionally, we develop a transformer-based Deep Learning (DL) model that provides suboptimal results with significantly reduced execution time. The proposed model achieves up to 96.13% of its MILP counterpart considering the packet admission metric while reducing the execution time by more than 99.98%.