The sixth generation mobile network (6G) is evolving to provide ubiquitous connections, multidimensional perception, native intelligence, global coverage, etc., which poses intense demands for network design to tackle the highly dynamic context and diverse service requirements. Digital Twin (DT) is envisioned as an efficient method for designing 6G that migrates the behaviors of physical nodes to the virtual space. However, in the high-dynamic 6G network, there still exist challenges in achieving accuracy and flexibility when constructing DT. In this article, we propose a Generative Artificial Intelligence (GAI)-driven mobile network digital twin paradigm, where the GAI is utilized as a key enabler to generate DT data. Specifically, GAI is capable of implicitly learning the complex distribution of network data, allowing it to sample from the distribution and obtain high-fidelity data. In addition, the construction of DT is closely related to various types of data, such as environmental, user, and service data. GAI can utilize these data as conditions to control the generation process under different scenarios, thereby enhancing flexibility. In practice, we develop a network digital twin prototype system to accurately model the behaviors of mobile network elements ( $i.e$ ., mobile users, base stations, and wireless environments) and to evaluate network performance. Evaluation results demonstrate that the proposed prototype system can generate high-fidelity DT data and provide practical network optimization solutions.

中文翻译：

---

用于移动网络的生成式人工智能驱动数字孪生


第六代移动网络（6G）正在向提供无处不在的连接、多维感知、原生智能、全球覆盖等方向发展，这对网络设计提出了强烈的要求，以应对高度动态的环境和多样化的服务需求。数字孪生 (DT) 被认为是设计 6G 的一种有效方法，可将物理节点的行为迁移到虚拟空间。然而，在高动态的6G网络中，构建DT时如何实现准确性和灵活性仍然存在挑战。在本文中，我们提出了一种生成人工智能（GAI）驱动的移动网络数字孪生范例，其中 GAI 被用作生成 DT 数据的关键推动者。具体来说，GAI 能够隐式学习网络数据的复杂分布，从而能够从分布中进行采样并获得高保真数据。此外，DT的建设与环境数据、用户数据、服务数据等各类数据密切相关。 GAI可以利用这些数据作为条件来控制不同场景下的生成过程，从而增强灵活性。在实践中，我们开发了网络数字孪生原型系统，以准确地对移动网络元素（即移动用户、基站和无线环境）的行为进行建模并评估网络性能。评估结果表明，所提出的原型系统可以生成高保真DT数据并提供实用的网络优化解决方案。