Deep reinforced traffic-aware CPU allocation in centralized RAN

The ongoing centralization of the Radio Access Network (RAN) and the higher Quality of Experience (QoE) requirements from next-generation services have significantly increased the computational demands of the Baseband Units (BBUs). These demands necessitate the efficient utilization of CPU resources for increased RAN performance. Contrary to the existing fixed CPU scheduling in BBU, this paper achieves dynamic CPU resource scheduling in BBUs by proposing Deep Reinforced CPU Allocation (DRCA) framework within RAN intelligent controller platform. By using RAN throughput as the feedback, DRCA learns to create a dynamic CPU resource schedule while taking several network state indicators into account. In particular, we propose three DRCA schemes, each focusing on a different network state indicator: Traffic-Aware DRCA (TA-DRCA), User-Aware DRCA (UA-DRCA), and Radio Resource-Aware DRCA (RA-DRCA). The impact of DRCA scheme is evaluated using network environment and state indicators from an industry-grade simulator and an open-source dataset. The results showcase 30% increase in packet processing throughput of BBU and up to 18% improved radio resource utilization achieved by the TA-DRCA on simulated datasets compared to the conventional static CPU allocation, highlighting the efficacy of DRCA framework in future cellular networks.