Getting ready for AI in the RAN

As networks evolve toward higher performance and efficiency, AI is rapidly becoming a critical tool to support and enable this transformation across the end-to-end infrastructure, including the RAN. With growing network complexity due to technological advances, a more open vendor ecosystem, increased traffic volume and more stringent QoE expectations, the RAN management and optimization have become unsustainable using the traditional, human-driven approach that has served us well for so long.

Operators have reached a tipping point where they struggle to manage their network and benefit from the expected performance increases, without a shift to automation powered largely by AI models. Continuing to manage this complexity manually is inefficient, error-prone, and costly.

Successfully adding AI to the RAN requires preparation and work. How can operators prepare to maximize the benefits that AI can bring to the RAN?

The cost of bad data

The success of AI-RAN significantly depends on the quality of the data fed into AI models. According to a recent Amdocs/IDC study, telecom operators currently leverage only around 50% of available data, primarily due to concerns about data quality and usability. Poor data quality has real and measurable costs in any environment.

In AI models, unreliable, unsuitable, inaccurate or inconsistent data not only reduces the effectiveness of AI (the garbage-in, garbage-out scenario), it amplifies the costs and distortion of bad data compared to a manual-driven environment. It can introduce biases, hallucinations and, more broadly, lead to flawed decisions that threaten network performance, security and profitability.

A solid foundation for AI-RAN: Data readiness

Regardless of AI model sophistication, benefits will not materialize without appropriate data. Preparation begins with data readiness, and it encompasses several critical aspects:

• Data pre-processing: Data must be cleaned, normalized, structured, and aggregated. It must be consistent across vendors, locations, and network elements and include legacy infrastructure. Preprocessing ensures that AI models draw upon accurate, consistent, and reliable data inputs.
• Data relevance: Operators must clearly identify and prioritize relevant data. Irrelevant or redundant data adds unnecessary complexity and computational costs without meaningful returns.
• Data granularity: Selecting the appropriate level of detail is crucial to get an AI model to extract the desired information and generate the appropriate response. Too much detail (e.g., an overly granular temporal resolution) may add noise and obfuscate trends; insufficient detail (e.g., from a cell site as a whole) may hide critical underlying drivers that cause a behavior.
• Data governance and sources: A clear governance structure is crucial to ensure data quality, consistency, and integrity, and to prevent costly inaccuracies, potential compliance issues, and security vulnerabilities.

Leveraging historical and real-time data

Preparing data is only the first step; operators must then integrate diverse sources according to the specific objectives of an AI model. This integration combines historical data, enabling learning from past experiences, with real-time data for current and predictive network optimization.

Data sources include:

• Topology: Information about network architecture, equipment deployment, and capabilities per location to understand interactions among network elements.
• Traffic: Historical and real-time traffic volumes, usage trends, temporal and geographic variability, and traffic patterns by application or service, to enable real-time and predictive optimization of traffic resources.
• User and device: Device types and capabilities, subscriber behaviors, and IoT connectivity, to deliver personalized services.
• Performance management: Real-time metrics on network performance and health, for predictive maintenance and reduction of downtime.
• Fault management: Data on alarms, failures, and maintenance records (both historical and real-time) to proactively identify and address network issues.
• External location and time-based data: External contextual data such as weather, major events, or geographic location impacting network usage patterns, to adapt the deployment and use of network resources based on the distribution of demand.

Benefits of integrating AI into the existing RAN infrastructure

Is AI integration in RAN worth the effort? Yes, it would be worth going through even without any plan to integrate AI. Having access to multiple sources of reliable data has always been beneficial. What is changing with AI is the greater ability to benefit from this data.

More specifically, however, AI can improve the RAN along multiple complementary directions: 

• Automation: Reducing manual tasks, speeding up fault detection and resolution, thus lowering operational costs and enhancing service quality.
• Optimization: Adjusting network parameters in real-time for optimal resource allocation, improved spectrum utilization, and enhanced user experience.
• Monetization of network data: Using AI-derived insights to create new revenue streams and innovative business models tied to network data.
• Capability exposure: Offering network functionalities to external developers and partners through APIs to stimulate innovation and facilitate new digital services.

Cashing on complexity

Injecting AI into the RAN is no longer optional; it’s imperative. With AI, operators can leverage the vast experience of their staff to fundamentally transform how networks are managed and optimized.

Preparation—ensuring AI models have access to high-quality, reliable, and relevant data from multiple sources—is essential for the successful integration of AI in the RAN.

Complexity is both a challenge and an opportunity. AI-RAN provides telecom operators with a powerful tool to convert network complexity into strategic advantage, enabling customized network optimization that directly enhances competitiveness and delivers measurable benefits.