Chris Pearson, President, 5G Americas (March 2024) – We live in the age of AI and machine learning, impacting every industry and platform. It seems like everywhere you go, AI/ML is appearing in every conversation and news article. This was a major storyline at Mobile World Congress 2024 in Barcelona, and is expected to continue to make more headlines this year and beyond.
Since this is a huge and new topic, we’re splitting this blog series into several articles that we’ll be publishing over the next few weeks. The article you are reading, Part I, focuses on AI efforts in 5G networks and how Generative AI is fundamentally different from other AI/ML processes. In Part II, we look at where Generative AI goes next in network planning and deployment, and spectrum management. Part III covers AI in content optimization and delivery, virtualization and network slicing, and user experience enhancement. Part 4 discusses AI/ML in security hardening and predictive maintenance. Finally, Part V explores AI/ML in training and simulation, as well as innovative services and applications.
Every wave of technological success seems to have a peak of excitement, followed by a trough of disillusionment, followed by a period of steady, productive success. AI/ML is no different, so it’s important to peel the onion and separate the hype from reality.
First, it’s important to recognize that the use of AI/ML is not new to wireless cellular networks. This has been a work in progress for several years, and 5G Americas has documented its evolution in a number of white papers. Recently, we have covered AI/ML in several important white papers, including 3GPP Technology Trends, Energy Efficiency and Sustainability in Mobile Communications Networks, and the State of Mobile Network Evolution.
In fact, the 3rd Generation Partnership Project (3GPP) has been working hard to establish specifications to further integrate AI/ML into 5G (and soon 5G-Advanced) networks. Wireless cellular networks are poised to take advantage of the latest advances in artificial intelligence, and great research continues to be completed.
Traditional uses of AI/ML in wireless cellular networks have tended to focus on a few key areas where intelligent classification and regression can help:
- Network optimization Self-organizing networks (SON) allow AI algorithms to dynamically adjust network parameters in real-time to improve performance and efficiency. Additionally, AI is being used in predictive maintenance to predict equipment failures and identify when maintenance is required before problems occur, minimizing downtime and increasing service reliability. You can improve your sexuality. For management orchestration and automation, 5G Americas explored the use of AI/ML in the automation and management of wireless cellular networks.
- Improved performance Through traffic prediction and management, ML models analyze traffic patterns to predict demand spikes and adjust network resources accordingly. It then automates resource allocation by dynamically allocating bandwidth and other network resources where they’re needed most, optimizing performance for high-demand applications like streaming, gaming, and virtual reality. In 5G Edge Automation and Intelligence, we explored how ML models can help improve the performance of 5G edge networks.
- Improved security Through anomaly detection, AI can monitor network traffic in real-time to detect and respond to anomalous patterns that may indicate security threats, such as DDoS attacks or unauthorized access attempts. Additionally, AI and ML can enhance security protocols, such as developing more secure biometric authentication methods and detecting vulnerabilities in network infrastructure.
- network slicing This network capability allows operators to leverage AI by allowing them to create multiple virtual networks with different characteristics on a single physical infrastructure. This is essential to support a wide range of applications, from IoT devices with low data needs to high-bandwidth applications such as 4K video streaming, which have specific requirements for latency, speed, and reliability. In Commercializing 5G Network Slicing, 5G Americas detailed how AI/ML can be used to improve network slicing technology and apply it to commercial use cases.
- Enhanced user experience. Even without network slicing, AI can be used to analyze network conditions and user behavior to dynamically adjust quality of service (QoS) settings and provide optimal service levels for different applications and services. It may be possible to secure it. AI is used in conjunction with predictive analytics to evaluate data about user behavior and device performance to predict user needs and adapt services accordingly to improve the user experience.
However, the advent of generative AI has revolutionized the application of machine intelligence. Generative AI is taking the industry by storm. First, what is generative AI? And how is it different from “regular” AI or machine learning?
Generative AI differs from traditional AI/ML in that it focuses on creating new, original content that mimics human-generated data, rather than simply interpreting or making predictions based on existing data. It is different from Generative AI utilizes models such as generative adversarial networks (GANs), variational autoencoders (VAEs), and transformers to learn the distribution of the underlying data and make it appear to be part of the original dataset. Creates a new instance. This is in contrast to traditional AI/ML, which primarily uses discriminative models for tasks such as classification and regression, and focuses on understanding patterns and making decisions based on input data. As a result, Generative AI enables innovative applications in content creation, data augmentation, and simulation, expanding the creative and practical potential of AI technology.
Introduced by Ian Goodfellow and his colleagues in 2014. Generative Adversarial Network (GAN) This represents a major advance in the ability of AI systems to generate realistic images. A GAN consists of her two neural networks, a generator and a discriminator, which are trained simultaneously through an adversarial process to produce highly realistic images and other types of data.
Around the same time as GAN, Variational autoencoder (VAE) It was developed as an alternative way to generate data. VAE is based on data encoding and decoding principles and provides a framework for learning deep latent variable models and generating new data samples.
Introduction of transformer model According to Vaswani et al. 2017 paved the way for significant advances in generative AI for natural language processing. Transformer models and subsequent iterations, such as GPT (Generative Pretrained Transformer) by OpenAI, introduced in 2018, have shown remarkable ability to generate coherent and context-relevant text, making them one of the most modern models in the natural language domain. is the basis of Generative AI.
Generative AI offers transformative potential for wireless cellular network operators by enabling advanced natural language processing capabilities, particularly through large-scale language models (LLMs). For example, wireless mobile phone network operators can leverage her LLM to create advanced chatbots for customer support, significantly improving the user experience. These chatbots understand and process customer queries in natural language and provide instant and accurate responses and solutions. One scenario might look like this:
scenario: Customers are experiencing connectivity issues with their mobile devices.
- Chatbot interaction:
- customer: “I can’t connect to the Internet with my mobile phone.”
- chatbot: “I’m here to help you. Let’s try a few steps. First, could you please make sure your mobile data is turned on? This is located under “Connections” in Settings. ”
- customer: “It’s powered on, but it still doesn’t work.”
- chatbot: “Okay. Try turning on “Airplane mode” and then turning it off. This may refresh the connection. ”
- customer: “It worked, thank you!”
Another, less obvious, but perhaps even more powerful example involves Large-Scale Language Models (LLMs), which are revolutionizing the way wireless cellular network operators manage and reprogram their network operations. . By integrating natural language processing capabilities, these models can interpret human commands and queries and translate them into actionable technical instructions and configurations. Here’s how it works in the context of reprogramming network operations:
scenario: Network operators need to reconfigure parts of their networks to improve performance, address congestion, or deploy new services.
process:
- Command input: Network operators provide commands in natural language, such as “Increase area 51 bandwidth allocation by 20% during peak hours to accommodate increased usage.”
- interpretation and translation: LLM interprets the intent of the command and translates it into a specific set of network configuration commands or scripts that the network management system can understand.
- automatic execution: These commands are automatically executed on the relevant network elements and adjust the configuration as directed without manual intervention.
- Review and feedback: Through the LLM, the system generates confirmation messages in natural language, such as “Area 51’s bandwidth allocation has been successfully increased by 20% during peak hours.”
These two examples just scratch the surface of what Generative AI can do for today’s wireless cellular networks. In the next part of our blog series, we will explore the potential impact of Generative AI on 5G and future wireless cellular networks as it relates to network planning and deployment, radio access network (RAN) configuration, and spectrum management.
-Chris
