With the public release of applications like ChatGPT, people have been able to leverage the power and potential of deep neural networks and machine learning (ML) to gain firsthand experiences. ChatGPT is a language model trained using massive amounts of text data from the internet and books, capable of generating text that sounds like it was written by a real person. This type of application perfectly demonstrates the advantages of artificial intelligence. It can continuously optimize its output in complex scenarios through a large amount of training data.
Wireless networks are inherently complex, generating massive amounts of data, and their complexity increases with each new generation of technology. These characteristics make artificial intelligence an ideal tool for optimizing wireless networks.
AI
With the maturity of 5G technology, AI and ML have been introduced into research by 3GPP (3rd Generation Partnership Project), which is an international standards organization that develops cellular technology standards. Currently, it is considering using artificial intelligence to improve the air interface, including network power saving, load balancing and mobility optimization. Since there are many potential use cases for the air interface, only a small subset has been selected for research in the upcoming 3GPP R18, covering channel state information (CSI) feedback, beam management and positioning. It should be noted that 3GPP has not developed artificial intelligence/machine learning models. Instead, it is trying to create a general framework and evaluation method to deploy artificial intelligence/machine learning models to different functions of the air interface [1].
Beyond 3GPP and the air interface, the O-RAN Alliance is exploring how to leverage artificial intelligence/machine learning to improve network orchestration and management. For example, the O-RAN Alliance architecture features a unique function called the RAN Intelligent Controller (RIC), primarily designed to assist AI and machine learning in optimizing various use cases. The RIC can manage both near real-time applications (xApps) and non-real-time applications (rApps). xApps for improving spectrum and energy efficiency, and rApps for leveraging AI for network orchestration and management, already exist. As the O-RAN ecosystem develops and matures, more xApps/rApps and applications optimized using RIC-based AI and machine learning will emerge.
Figure 1: ORAN Network
6G
While 6G is still in its early stages, it's certain that artificial intelligence/machine learning will become a fundamental component of all aspects of future wireless communication systems. At the network level, although there's no formal definition, the term "AI-native" is already widely used in the industry. One way to observe these AI-native networks is to infer the above diagram (Figure 1) based on the current virtualization technologies and de-aggregation trends of the RAN (Radio Access Network). Each block in the network may contain AI/machine learning models, which may vary between different vendors and applications (Figure 2).
Figure 2: ORAN 6G Network
AI-native networks can also refer to networks built to run native artificial intelligence/machine learning models. Please refer to the design flow below (Figure 3). In traditional 5G networks, the air interface consists of different parts, each designed by humans. In 5G-Advanced networks, each part will utilize machine learning techniques to optimize specific functions. In 6G networks, the entire air interface may be designed by artificial intelligence using deep neural networks.
Figure 3: From integration with AI to the development of AI-native networks [2]
AI
Drawing inspiration from the idea that artificial intelligence (AI)/machine learning can be used to improve network orchestration and management, 6G hopes to leverage AI and machine learning to address optimization challenges. For example, AI can turn components on and off based on real-time operational data to reduce overall network power consumption. Currently, xApps and rApps achieve this at the base station level by turning on and off high-energy-consuming components such as power amplifiers that are not in operation. However, AI's ability to quickly solve challenging computational problems and analyze massive amounts of data makes it possible to optimize network performance on a larger scale, even citywide or nationally. It allows for the shutdown of entire base stations during periods of low usage, and cell reconfiguration to meet users' real-time needs in a green, low-carbon, and energy-efficient manner using as few resources as possible. Currently, it's not possible to reconfigure base stations and entire city networks in this way; reconfiguring and testing any changes to network configuration typically takes days or weeks. Nevertheless, the development prospects of various AI technologies are vast, and they remain a primary consideration for infrastructure providers.
Summarize
The application of artificial intelligence (AI) in wireless networks will not wait until the advent of 6G networks. The entire ecosystem is actively researching and developing new models to integrate into existing and future wireless communication systems. However, these models are still new and require evaluation for rigor and reliability. Properly training AI models on diverse datasets, quantifying their improvements over traditional techniques, and defining new testing methods for AI-driven modules are all crucial steps that must be taken with the adoption of new technologies. As AI models and testing methods and technologies mature, there is no doubt that AI will revolutionize the wireless communication industry within the next 5-10 years.
