What is artificial intelligence-radio access network (AI-RAN)?

How is AI-RAN different from traditional RAN?

A conventional radio access network is a vital part of existing cellular networks. It uses radio signals to exchange data, ensure command and control, manage resources and handle roaming behaviors between cellular locations. A RAN relies on traditional radio devices -- such as transmitters, receivers and antennas -- to provide radio communication between devices such as smartphones. RANs have evolved from 1G to modern 5G communication, which can now support virtualization, cloud computing and other advanced technologies.

AI-RAN adds AI capabilities to the RAN infrastructure. This is typically accomplished at the underlying computing level with a cloud-centric computing approach. The compute infrastructure is completely unified and general-purpose -- there is no RAN-specific hardware. The result is a containerized, highly scalable and multi-tenant computing environment that can run cellular workloads and AI workloads concurrently. Similarly, AI-RAN platforms operate in a fully software-defined approach that can provision resources, launch and manage applications, and handle optimizations to help ensure top performance of both cellular and AI workloads. AI-RAN is intended to align with Open RAN (O-RAN) principles to provide a flexible and interoperable environment.

The various ways in which AI is combined with existing cellular capabilities -- AI for RAN, AI on RAN, and AI and RAN -- offer the potential for new services. The following table provides a brief comparison between conventional RAN and AI-RAN management.

Components of AI-RAN

The principal components of AI-RAN include the infrastructure, the AI and data running on the infrastructure, and the orchestration and management systems that keep everything organized.

Infrastructure

AI-RAN relies on a single common computing infrastructure that can run traditional RAN and advanced AI workloads. This unified computing infrastructure typically consists of commercial off-the-shelf (COTS) servers. COTS servers eliminate specialized hardware and enable the flexibility to run different workloads in various situations.

The AI-RAN also includes varied hardware acceleration platforms optimized for specific computing tasks, such as the following:

• CPUs for general-purpose tasks and application control.
• Graphics processing units (GPUs) for accelerating signal processing tasks as well as AI and ML processing.
• Data processing units (DPUs) for critical network and data processing tasks, which are well suited to real-time or latency-sensitive applications.

Infrastructures should be multi-tenant and multipurpose -- designed to support any RAN workload, any AI workload and any cloud-native network function. The infrastructure should also be highly scalable and interchangeable, enabling deployment as centralized-RAN, distributed-RAN or massive multiple-input, multiple-output using the same infrastructure components.

AI and data

AI-RAN systems incorporate AI and ML models, along with data and data pipelines. This AI layer provides the intelligence of the AI-RAN system and makes AI workloads run efficiently. The AI layer enables the following:

• Data collection to gather real-time data from equipment.
• KPIs such as latency.
• Radio measurements such as signal strength.

The AI models process data and provide insights for improving radio system performance using advanced techniques such as dynamic spectrum allocation. Similarly, data can be imported from varied sources and used to drive AI applications for advanced services within the AI-RAN environment.

This AI layer also includes all the AI libraries, tools and frameworks -- such as the Nvidia Aerial CUDA-Accelerated RAN -- needed to build, train, deploy and monitor AI-RAN capabilities. Some systems employ digital twin capabilities, providing virtual replicas that simulate the actual environment for testing and validation prior to AI model deployment.

Orchestration and management

The orchestration and management function handles AI-RAN automation -- often referred to as an intelligent orchestrator or RAN intelligent controller. It can also provision and manage the computing resources and network functions needed to run RAN and AI workloads.

Orchestration also handles workload scheduling to ensure that AI and RAN workloads run concurrently on the common infrastructure while meeting performance requirements. AI management systems also learn from data and analytics, enabling AI management to improve and adapt to changing environments and requirements over time.

Why AI-RAN matters to enterprise network strategy

AI-RAN enables the cellular network to become more efficient, intelligent, adaptable and directly suited to running enterprise AI workloads as well as moving huge amounts of data within the RAN itself. AI-RAN can enhance cellular networks to provide the immense bandwidth and low latency required for AI workloads.

Ultimately, enterprise users who engage AI-driven RAN cellular providers can realize competitive advantages over conventional RAN systems. Enterprise adopters should consider the following implications of AI-RAN on their network strategy:

• Network optimization. AI-RAN system infrastructures can process data close to the network edge, where raw data is generated and collected. This can reduce bandwidth demands when moving huge data sets to a centralized location, such as the public cloud, for processing. Other optimizations include dynamic resource allocation, using AI to predict RAN traffic demands and proactively allocating resources such as radio spectrum utilization, network bandwidth and underlying computing power to maintain AI application performance.
• System reliability. Using AI to help collect and analyze RAN performance data can help providers deliver proactive management and maintenance for greater RAN availability and resilience. Predictive maintenance identifies potential faults and directs corrective actions before disruptions occur. AI-RAN systems can detect anomalies and reconfigure networks to remediate outages, providing a level of self-healing and reduced downtime for RAN users and applications.
• New capabilities. AI-RAN offers services beyond basic cellular connectivity. These include edge computing services such as on-site computer vision/recognition and autonomous drone inspections. Virtualization and intelligent orchestration can work with AI provisioning to support dynamic network segments -- a technique known as network slicing -- for running demanding enterprise applications such as medical, industrial or financial applications.
• Security and risk management. AI-RAN supports greater data and application security in the RAN itself with real-time anomaly and threat detection that can identify potential risks before local tools do. Processing sensitive data at the edge enables AI-RAN to help organizations support data privacy and data sovereignty demands to maintain regulatory compliance. The role of AI in the RAN also helps the provider learn and adapt its security to potential new threats in real time.

Challenges and considerations for AI-RAN adoption

The creation of an AI-RAN environment is a deliberate business choice made by cellular providers. As with so many other advanced technologies, enterprise users should consider a variety of factors when making important strategic decisions regarding AI-RAN adoption. Common challenges and considerations for AI-RAN adoption include the following:

• End-to-end infrastructure. Cellular providers don't exist in a vacuum. Their AI-RAN capabilities work best for users with a consistent, cloud-native, software-defined infrastructure that is fundamentally compatible with O-RAN platforms suited to flexible and scalable AI workloads. Organizations should evaluate their own network and services carefully to see how well workloads and data can be exchanged with AI-RAN systems. Improvements to enterprise infrastructure can bring performance and reliability improvements outside of the cellular provider.
• Evaluate traffic demands. AI depends on enormous volumes of data, both real-time and historical. Gathering, pre-processing, validating, storing and moving volumes of data across a network -- especially mobile networks affected by real-world limitations -- will dramatically change traffic patterns and loads within the enterprise local area network, mobile devices and the AI-RAN provider. Data management issues will affect the performance of the AI application, so understanding traffic patterns and volumes can be critically important for enterprise infrastructure planning.
• Focus on data quality. Enterprise AI projects depend on high-quality data that is complete, accurate, consistent and relevant. This demands careful attention to data quality issues and management. Consider the requirements involved in producing, validating and maintaining high-quality data, as well as the implications of moving that data across a cellular network, where a provider's AI services can deliver enterprise value. Without quality data, no AI system will operate correctly for very long.
• Start small. Using AI-RAN systems for enterprise projects is not an all-or-nothing proposition. Test and develop AI-RAN skills with small scale, highly-targeted use cases that pose a low risk for the business but can demonstrate measurable return on investment, such as in edge computing cases. These vital test cases provide important opportunities to collaborate with AI-RAN vendors, gain expertise and build skills that can all be applied to more complex and impactful AI projects later.
• Consider data governance. Whether applied locally, in a public cloud or through a cellular AI-RAN, the data gathered and used for AI applications presents serious concerns for data security and compliance, such as protecting personally identifiable information. Organizations should evaluate data security, access, retention and destruction, as well as implement a comprehensive incident management protocol.
• Look for monitoring and management. AI-RAN provides opportunities for enterprise users to employ AI services and develop AI applications that can run within a cellular provider's network. This must also include a consistent and compatible enterprise infrastructure. Just as an enterprise must monitor external services such as SaaS, the ability to observe and monitor an end-to-end AI-RAN environment -- which includes the local enterprise infrastructure -- can be crucial for optimizations, quality of service and reliability.

Enterprise use cases of AI-RAN

AI-RAN supports both wireless connectivity and AI applications across a low-latency, high-throughput 5G or faster wireless network. This offers a suitable environment for AI workloads that emphasizes remote operations and data gathering, including large-scale industrial automation, metropolitan area automation -- such as in public safety -- and real-time analytics. Consequently, AI-RAN provides many of the same enterprise use cases found in more conventional AI contexts.

• Healthcare support. The low latency, high reliability and high throughput of modern cellular systems enables AI-RAN to support advanced healthcare tasks such as real-time surgical assistance, remote patient monitoring and alerting, and medical image analytics.
• Logistics. AI can analyze sensor data from varied industrial equipment, vehicles and storage locations to manage and optimize logistics. Enterprises can use AI-RAN infrastructures across the supply chain to help forecast product demand, manage inventory, organize manufacturing schedules and plan deliveries.
• Predictive maintenance. IoT devices can deliver real-time sensor data for AI analysis across the AI-RAN network. AI workloads can detect data anomalies and predict possible failures -- even automating the scheduling and deployment of maintenance resources to mitigate disruption.
• Predictive vehicle diagnostics. In automotive applications, AI-RAN can collect vehicle data for AI analysis. Vehicle manufacturers can then use analytics to predict possible service issues and alert the driver to seek maintenance and minimize breakdowns or more costly vehicle failures.
• Quality control. Cameras can employ computer vision to stream images across the RAN to an AI application that evaluates each product for defects or other quality parameters. This helps bolster product quality at the edge and eliminates the need to send massive amounts of data to a central point for analysis.
• Retail experience. AI-RAN can integrate with smart displays and send special offers to shoppers' mobile devices in response to shopping behaviors and interests. Similarly, virtual AI assistants can help guide customers to products of interest and answer their questions. In addition, AI-RAN can use edge computing to perform shopper flow analysis to understand customer movement and optimize store layouts for better shopping experiences.
• Robotics. AI-RAN can use the low latency of wireless networks to receive real-time sensor data from autonomous systems such as robots and autonomous vehicles to guide them around complex and dynamic industrial areas.
• Security and surveillance. Enterprises and public safety organizations can use AI-RAN to collect and analyze real-time video streams from cameras. AI analysis can then alert organizations to potential threats or security breaches -- even calling local law enforcement to address the potential issue.
• Traffic management. Municipal governments can use AI-RAN systems to collect and analyze IoT, video and traffic control system data. The resulting analyses can then be used to optimize traffic control patterns and manage traffic flows for more efficient transportation routes and lower congestion in busy areas.

AI-RAN and the future of Open RAN

AI-RAN and O-RAN are deeply interrelated concepts. AI-RAN represents the integration of AI into radio access networks. This idea advances the fundamental concept of O-RAN, which focuses on standardized multivendor architectures to support openness and interoperability.

In effect, O-RAN is the foundation for AI-RAN, which brings a new layer of intelligence to RAN systems. AI-RAN will affect O-RAN by facilitating the development and deployment of AI applications on the RAN, such as AI-native networks, acting as a key building block for future 6G communication services.

Should telecommunications companies rent GPUs from a hyperscaler or buy them for AI-RAN? Learn more about the AI-RAN landscape and competing opinions on the still-evolving technology.