With the rapidly increasing number of bandwidth-intensive terminals capable of intelligent computing and communication, such as smart devices equipped with shallow neural network models, the complexity of multiple access for these intelligent terminals is increasing due to the dynamic network environment and ubiquitous connectivity in 6G systems. Traditional multiple access (MA) design and optimization methods are gradually losing ground to artificial intelligence (AI) techniques that have proven their superiority in handling complexity. AI-empowered MA and its optimization strategies aimed at achieving high Quality-of-Service (QoS) are attracting more attention, especially in the area of latency-sensitive applications in 6G systems. In this work, we aim to: 1) present the development and comparative evaluation of AI-enabled MA; 2) provide a timely survey focusing on spectrum sensing, protocol design, and optimization for AI-empowered MA; and 3) explore the potential use cases of AI-empowered MA in the typical application scenarios within 6G systems. Specifically, we first present a unified framework of AI-empowered MA for 6G systems by incorporating various promising machine learning techniques in spectrum sensing, resource allocation, MA protocol design, and optimization. We then introduce AI-empowered MA spectrum sensing related to spectrum sharing and spectrum interference management. Next, we discuss the AI-empowered MA protocol designs and implementation methods by reviewing and comparing the state-of-the-art, and we further explore the optimization algorithms related to dynamic resource management, parameter adjustment, and access scheme switching. Finally, we discuss the current challenges, point out open issues, and outline potential future research directions in this field.

翻译：随着具备智能计算与通信能力的带宽密集型终端（例如搭载浅层神经网络模型的智能设备）数量迅速增长，6G系统中动态的网络环境和泛在连接使得这些智能终端的多址接入复杂性日益增加。传统的多址接入设计与优化方法在处理复杂性方面逐渐让位于已证明其优越性的人工智能技术。旨在实现高服务质量的人工智能赋能多址接入及其优化策略正受到越来越多的关注，特别是在6G系统中对时延敏感的应用领域。本文旨在：1）阐述人工智能赋能多址接入技术的发展与比较评估；2）聚焦于人工智能赋能多址接入的频谱感知、协议设计与优化提供及时综述；3）探索人工智能赋能多址接入在6G系统典型应用场景中的潜在用例。具体而言，我们首先通过整合频谱感知、资源分配、多址接入协议设计与优化中各种有前景的机器学习技术，提出面向6G系统的人工智能赋能多址接入统一框架。随后介绍与频谱共享及频谱干扰管理相关的人工智能赋能多址接入频谱感知技术。接着，通过回顾和比较现有先进成果，探讨人工智能赋能多址接入协议设计与实现方法，并进一步深入研究动态资源管理、参数调整和接入方案切换相关的优化算法。最后，我们讨论当前面临的挑战，指出开放性问题，并展望该领域未来潜在的研究方向。