通信,拉近世界的距离!
您的位置:首页 >资讯 >业界新闻 >查看资讯

Omdia 调研:AI RAN 从实验室走向商用,12% 运营商已启动部署

移动通信网 程然

AI 模型直接部署在无线基站内部,正从概念验证走向商用现网。Omdia 最新发布的 AI RAN 运营商调研显示,已有 12% 的受访运营商完成部署或制定了明确部署计划,另有 27% 正在现网开展试点。这组数据表明,RAN 系统设计与性能优化正站在一场深层变革的门槛上——业内称之为 AI RAN 或 AI 原生 RAN。

从算法优化到 AI 原生:基站侧的范式转换

在无线接入网中,将 AI 模型直接嵌入基站的 Layer 1/2/3 协议栈,被视为 AI 在 RAN 领域最基础也最具挑战性的应用方向。其核心目标在于突破传统算法在实时无线过程中的性能天花板,在覆盖、能耗、上行性能和频谱效率等多个维度提升系统表现,并针对不同场景优化用户体验。与推翻重建不同,这类 AI 模型多以软件升级形式加载到存量 RAN 设备上,部署门槛相对可控。

具体用例已经相当丰富:数字预失真、信道估计、AI 管理的波束赋形、MU-MIMO 用户配对、AI 优化的链路自适应以及多层资源协调等,均已成为设备商和运营商竞相验证的方向。更值得关注的是模型组合的潜力——单个模型的增益叠加后,有望带来系统整体性能的跃升。目前多个模型能否同时运行、模型间交互的动态特性如何,仍是业界探索的前沿课题。

【关键数据】

- 已部署/计划部署运营商占比: 12%

- 现网试点占比: 27%

- 实验室测试/桌面研究占比: 分别为 27% 和 30%

- 全球无线基站总量: 超过 1000 万个

- 已明确的基站侧 AI 用例: 6 大类

场景化定制与规模化挑战并存

AI 模型的另一价值在于场景化定制能力。针对不同运营商网络、不同地理位置乃至不同垂直场景训练专属模型,正在成为可能——农村小区边缘的最优策略,与高密度城区、铁路或高速公路沿线的需求可能截然不同。这种精细化调优能力,是传统统一算法难以企及的。

不过,调研数据同样揭示了现实的另一面。处于实验室测试阶段的运营商占比为 27%,仍在桌面研究阶段的达 30%,两者几乎平分秋色。这意味着,要在全球超过 1000 万个基站中实现 AI 模型的广泛而深入的部署,仍需相当长的时间周期。业界需要警惕过度炒作,扎实解决模型泛化、验证标准和运维复杂度等工程化难题。

单站多能:AI RAN 搭上 6G 平台化快车

从长期看,AI RAN 的处理需求将持续攀升。随着运营商利用 giga-MIMO 等先进射频技术部署新频谱,并将 4G/5G 与未来 6G 整合到统一基站平台上,对计算资源的要求水涨船高。同一平台还需支撑 ISAC(通感一体化)等新业务,以及向 RAN 域延伸的 AI 辅助工作负载。基站正在从单一连接设备演变为集连接、感知、计算于一体的智能平台。

Omdia 此项调研旨在帮助产业生态更清晰地理解 AI RAN 将如何改变网络的构建、运营和变现方式。对设备商和运营商而言,谁能率先跑通基站侧 AI 模型的规模化部署路径,谁就能在下一轮网络智能化竞争中占据主动。


出处:Operator Survey: Real progress toward AI RAN

英文原文
Applying AI models directly in the radio base station is one of the most fundamental and demanding uses of AI in radio access networks. It is the starting point for what is known as AI RAN or AI native RAN. The aim is to improve real-time radio processes beyond what is achievable with conventional algorithms to increase RAN system performance and optimize user experiences in different scenarios. According to a new survey from Omdia, operators are on the verge of a substantial, far-reaching change in RAN system design and performance. The Omdia AI RAN Operator Survey asked operators about their plans to deploy AI models directly in base stations (i.e., in the Layer 1/2/3 radio stack) in live commercial networks. The figure below shows that a small lead group of respondents (12%) have already deployed or plan to deploy, and that close to a quarter (27%) are in live network trials. AI models deployed as software updates to existing RAN equipment can improve many aspects of RAN system performance, including coverage, energy consumption, uplink performance and spectral efficiency. Example use cases for AI models in radio base stations include digital pre-distortion, channel estimation, AI-managed beamforming, MU-MIMO pairing, AI-optimized link-adaptation and multilayer resource coordination. What is perhaps most exciting is that the initial deployments are just the first of many AI models. How many models can run simultaneously and the dynamics of their interactions remain areas for exploration. But the potential is to combine the gains from individual models to radically improve overall system performance. There is also the opportunity to deploy models trained for different operator networks or locations, as well as specific use cases. What works best for rural cell edge may differ from what works in dense urban locations, which may differ from what works for railways or roads. It is important not to get too overhyped. With roughly equal splits between those in lab trials (27%) and those in the desk research phase (30%), the survey also shows it will take time for AI models to be extensively and deeply deployed in the world's 10+ million radio base stations. As operators deploy new spectrum with advanced radio technologies such as giga-MIMO and consolidate 4G/5G and 6G onto a single base station platform, the processing requirements for AI RAN will increase. The same platforms should also support new services such as ISAC and AI-assisted workloads to the RAN domain. The Omdia AI RAN Operator Survey, published here on Light Reading, is designed to help the technology ecosystem better understand how AI RAN will change how networks are built, operated and monetized. Download a free copy [here](https://lr-resources.lightreading.com/free/w_defa12109/prgm.cgi?a=1). Gabriel leads mobile network research for Omdia. His coverage includes system architecture, RAN, core, and service-layer platforms. Key research topics include 5G, open RAN, mobile core, and the application of cloud technologies to wireless networking. Gabriel has more than 20 years' experience as a mobile network analyst. Prior to joining Omdia, he was chief analyst for Light Reading's Insider research service; before that, he was editor of IP Wireline and Wireless Week at London's Euromoney Institutional Investor.
微信扫描分享本文到朋友圈
扫码关注5G通信官方公众号,免费领取以下5G精品资料
5G通信公众号
  • 1、回复“YD5GAI”免费领取《中国移动:5G网络AI应用典型场景技术解决方案白皮书》
  • 2、回复“5G6G”免费领取《5G_6G毫米波测试技术白皮书-2022_03-21》
  • 3、回复“YD6G”免费领取《中国移动:6G至简无线接入网白皮书》
  • 4、回复“LTBPS”免费领取《《中国联通5G终端白皮书》》
  • 5、回复“ZGDX”免费领取《中国电信5GNTN技术白皮书》
  • 6、回复“TXSB”免费领取《通信设备安装工程施工工艺图解》
  • 7、回复“YDSL”免费领取《中国移动算力并网白皮书》
  • 8、回复“5GX3”免费领取《R1623501-g605G的系统架构1》
  • 本周热点本月热点
    通信人才网

    最热通信招聘

    更多