As 5G demands continue to rise, Radio Access Networks will benefit from being automated by AI and ML.
Registered users can unlock up to five pieces of premium content each month.
Log in or register to unlock this Insight.
3GPP Set AI/ML in the Framework for RAN and Air Interfaces
At the end of 2021, 3GPP approved the Release 18 work package, the fourth 5G release, and set a start for 5G Advanced. One of the noticeable topics is Artificial Intelligence (AI)/Machine Learning (ML) for Radio Access Network (RAN) and air interface, starting an evolutionary path to the full automated network using AI/ML, including the user equipment (UE) side.
While the long-term target of a full automated network is challenging, 3GPP breaks down the implementation step by step. Firstly, it will look at the essential framework in RAN for AI/ML applications such as data collection enhancements and signaling support for AI/ML-based network energy saving, load balancing, and mobility optimization. Second, it will explore AI/ML techniques to improve channel state information (CSI) feedback, beam management, and positioning for the air interface. Meanwhile 3GPP aims to standardize interfaces and platforms and let the market decide about algorithms.
Competition has Started Among Major Vendors
AI/ML is not a new concept, though in recent years it has become more and more popular in implementation. In the wireless world, 3GPP started addressing AI/ML features in Release 17 and will work to establish the framework from now on, while the Telecom Infra Project (TIP) OpenRAN 5G NR Project Group in 2020 also launched a new subgroup RAN Intelligence & Automation (RIA) to leverage AI/ML into the RAN market.
Major vendors with operators have launched several tests to prove AI/ML strength in network optimization. For example, Nokia worked with China Mobile to complete live trials of an AI-powered RAN to forecast bandwidth and detect anomalies with 90% accuracy early in 2021. Huawei launched an AI-based intelligent solution with AIS Thailand with increasing network capacity by 15%. Ericsson embedded AI in the RAN compute software and achieved a 15% reduction in energy OPEX. Qualcomm has been developing On-device AI solutions to enable optimization between devices and RAN functions, such as CSI feedback and network loading, which are in line with 3GPP Release 18 objectives.
Network vendors and operators have advantages in deploying AI/ML solutions for the 5G RAN compared to the third-party AI/ML developers because they have advanced knowledge in network architectures and have proprietary hardware. Additionally, network operators have ownership of user traffic data which is a great asset for AI/ML model training.
Synergy Between 5G and AI/ML
The introduction of AI/ML is driven by predictive rapid growth in 5G network usage. First, the data traffic will keep growing for the consumer market, reaching 3,115 exabytes in 2026 from 731 exabytes in 2021 with 34% CAGR over the period, according to ABI Research. Network operators need to significantly improve spectral efficiency and deploy more infrastructure to handle such a demand. Meanwhile, power consumption and deployment cost need to be controlled under a profitable level which calls for systematic network energy saving and smarter deployment decisions. Moreover, the complexity of the network optimization will be hard to manage with current approaches limited with presumed models. Alternatively, 5G is designed with flexible network Key Performance Indicators (KPI) configurations for various UE scenarios, which is with non-linearity, hard to be modeled, and is exactly the strength of AI/ML.
As mentioned above, developments from major vendors have shown a good start of AI/ML development in network optimization. However, further progress is expected for the future 5G-Advanced market, given expected demand in data traffic, spectral efficiency, energy saving, and the complexity in network management. Network vendors need to continue the development in AI/ML applications across the whole network and test with operators for predictive accuracy under diversified use case scenarios. Meanwhile, network operators need to balance their investment in computing assets for the best cost efficiency as the AI/ML applications require intensive computing power support.
AI/ML-based solutions can have a much shorter time-to-market due to their software dominant nature, which gives another reason to shift innovation focus to AI/ML applications to improve the 5G network system. Major vendors have started to build AI/ML into their baseband units for RAN compute software and deployed with a few operators. With more successful trials, ABI research expects a fast launch of AI/ML applications in 5G mobile network across the world firstly focusing on areas in 3GPP Release 18, such as energy saving, beam management to extend coverage, and traffic loading balance. Different from a gradual coverage of hardware installment, a wide coverage will be reached with a preliminary version then being kept agile for frequent upgrades.