@

초록

Emerging Consumer Electronics (CE) systems, such as connected vehicles, immersive AR/VR environments, cloud gaming platforms, and intelligent home robotics, are increasingly characterized by their dependence on distributed intelligence, seamless connectivity, and autonomous control. These systems often span multiple Mobile Network Operators (MNOs) and serElectronicsders, requiring coordinated management of communication, computing, and control resources at the network edge. When AI and RAN workloads share edge infrastructure, this coexistence naturally achieves integration: RAN workloads provide communication and control while AI workloads supply computing intelligence. However, as MNOs increasingly share infrastructure under the Open Radio Access Network (O-RAN) paradigm, efficient resource allocation across heterogeneous workloads and independent Service Level Agreements (SLAs) becomes critical. At the same time, preserving consumer data privacy across administrative boundaries remains a major concern. This paper proposes the Converged AI and O-RAN Architectural (CAORA) framework, a federated and SLA aware orchestration solution that integrates AI and RAN workloads with communication, computing, and control. CAORA employs local spike-aware workload forecasting using attention-based LSTMs and aggregates predictions through federated learning to avoid raw data sharing between operators. A centralized Soft Actor Critic (SAC) reinforcement learning agent then allocates edge resources fairly and efficiently while meeting SLA constraints. Experiments on real 5G traffic traces from Barcelona show that CAORA achieves about 90% task completion, reduces SLA violations to below 10%, and reaches nearly 100% resource utilization under peak load.

키워드