6G内生智能与信道基础模型

摘要：人工智能（AI）与通信系统的深度融合已成为6G的关键目标与核心标志之一，内生智能（Native AI）被普遍视为6G系统的重要特征。阐述了对6G内生智能内涵与需求的理解，在此基础上系统梳理了无线通信领域AI研究范式的演进历程，揭示了基于监督学习的传统AI模型在支撑6G内生智能方面存在的固有局限。针对上述挑战，提出了信道基础模型（CFMs）的概念框架，系统介绍了其预训练方法体系及面向各类信道相关任务的任务适配机制。认为6G内生智能需具备强大的任务适应性与场景泛化能力，而信道基础模型凭借其核心技术特征，有望成为未来6G内生智能的关键技术选项之一。

关键词：6G内生智能；信道基础模型；掩码重建；对比学习；通感一体化；自监督学习

Abstract: The deep integration of artificial intelligence (AI) and communication systems has emerged as a key objective and core hallmark of 6G, with native AI being widely recognized as an essential characteristic of 6G networks. An understanding of the connotation and requirements of 6G native intelligence is first elaborated. Based on this, the evolution of AI research paradigms in wireless communications is systematically reviewed, revealing the inherent limitations of traditional supervised learning-based AI models in supporting 6G native intelligence. In response to the above challenges, a conceptual framework of channel foundation models (CFMs) is proposed, and its pre-training methodology as well as task adaptation mechanisms for various channel-related tasks are systematically introduced. It is envisioned that 6G native intelligence requires strong task adaptability and cross-scenario generalization capabilities, and channel foundation models, by virtue of their core technical features, are expected to become one of the key technological enablers for future 6G native intelligence.

Keywords: 6G native AI; channel foundation models; masked channel modeling; contrastive learning; integrated sensing and communication; self-supervised learning