To address the limitations of current radio access networks (RANs), centralized RANs adopting the concept of flexible splits of the BBU functions between radio units (RUs) and the central unit have been proposed. This concept can be implemented combining both the Mobile Edge Computing model and relatively large-scale centralized Data Centers. This architecture requires high-bandwidth/low-latency optical transport networks interconnecting RUs and compute resources adopting SDN control. This paper proposes a novel mathematical model based on Evolutionary Game Theory that allows to dynamically identify the optimal split option with the objective to unilaterally minimize the infrastructure operational costs in terms of power consumption. Optimal placement of the SDN controllers is determined by a heuristic algorithm in such a way that guarantees the stability of the whole system. Finally, multi-agent learning methods were investigated in order to expand the model to more sophisticated scenarios where many RUs with limited information are interacting.

为了解决当前无线电接入网（RAN）的局限性，已经提出了采用在无线电单元（RU）和中央单元之间的BBU功能的灵活划分的概念的集中式RAN。可以结合移动边缘计算模型和相对大型的集中式数据中心来实现此概念。这种架构要求高带宽/低延迟的光传输网络互连RU，并采用SDN控制来计算资源。本文提出了一种基于进化博弈论的新颖数学模型，该模型允许动态识别最佳拆分方案，其目的是就功耗方面单方面最小化基础设施的运营成本。SDN控制器的最佳位置由启发式算法确定，以确保整个系统的稳定性。最后，研究了多智能体学习方法，以将模型扩展到更复杂的场景，在这些场景中，许多具有有限信息的RU相互作用。