Artificial intelligence (AI) has already improved efficiency and reliability across a variety of industries, and is predicted to have a similar impact on cellular technology. In fact, AI market revenue in the global telecommunication industry is expected to reach $19.5 billion by 2030 — nearly ten times its scale in 2023.
We see AI integration with RAN as the first steps toward a future where AI is an inherent part of radio networks, allowing communication service providers (CSPs) to continuously adapt to simultaneously meet specific user experience and energy efficiency goals — cost-effectively, reliably, and sustainably.
AI learnings from 4G help advance 5G features
By monitoring traffic patterns to predict user demand, AI dynamically adjusts the network’s radio link configurations, resulting in an 11.6% throughput improvement for heavy users, and a 50% increased cell edge download (DL) throughput — measured in the field by CSPs.
Since AI can anticipate the best target for handover due to poor coverage, user movement and network congestion, Machine Intelligence Enabled Mobility has achieved a 1.2% reduction in overall drop rate, and a 10.9% reduction in inter-frequency handover failure, creating measurable improvements in user experience and reliability.
AI also monitors traffic, optimizing energy consumption by 14% per radio site without compromising performance.
All these AI-powered features amount to a significant improvement in network performance while also delivering significantly improved RAN energy efficiency.
New AI opportunities in 5G
5G brings edge computing technology to a far larger variety of devices than 4G while reducing the processing and energy consumption load previously placed on those devices. In turn, this affects even more opportunities for AI technology to optimize reliability, speed and efficiency.
Intelligent cell shaping, a process by which the beam shape of the radio signal is adapted in real time, is one way this is coming to life. AI plays a role by boosting improvements at a network level through centralized automation – essentially learning and adjusting along the way. For example: By adjusting the tilt (vertical plane) and horizontal beamwidth of the Active Antenna Systems (AAS) autonomously as traffic distributions change, we’ve seen a 5% increased throughput in downlink and a 30% increased throughput in uplink — a 35% increased cell edge speed overall.
AI’s potential for a high return on investment (ROI) in the RAN
The great impact of generative AI, when compared to the older discriminative AI, is the ability to generate new data in the form of text, image, audio and video. In the context of the RAN, CSPs gain the ability to create far more sophisticated data predictions, and then execute strategic decisions based on those predictions. Maximizing the impact of generative AI’s utilization in the RAN must be distilled into three categories: unique use cases, high complexity use cases, and multi-objective use cases.
- In unique use cases, specific AI tools will utilize prediction, anomaly detection, and analysis to predict demand and challenges and optimize performance.
- In high complexity use cases, like link adaptation features, AI will consider multiple factors to determine the best signal modulation for optimal transmission.
- And in multi-objective use cases, AI will seek an optimal solution under complex constraints and trade-offs — for example, maintaining network key performance indicators (KPIs) and maximizing energy savings while handling handover choices.
Challenges (and solutions) in implementing AI in the RAN
Integrating any new technology can be challenging from a business and operations perspective. To successfully implement AI in RAN, CSPs should focus on creating a coherent strategy with clear, achievable targets that are supported across the company.
Having a strong strategy in place can guide CSPs through AI implementation — but that only works when technical challenges are also addressed.
Data management challenges result from the massive amounts of data generated by the rapidly-growing network of users utilizing 5G technology from the wide variety of devices in the network. Managing that data efficiently will be paramount to the success of these projects. As AI technology continues to quickly evolve, CSPs also need to consider future-proof tools and architecture for optimal lifecycle management.
CSPs also need to consider the hardware challenge: due to the exponential growth of AI processing needs, hardware must be designed with capabilities like storage and processing that is optimized for AI model execution.
Actionable Insights
We have identified three preliminary steps to take towards activation in this space: modernization of networks, simplifying operations, and data driven organizational transformation through vendor partnerships.
- Simplifying operations, Distributed AI monitors can break the operational complexity curve with their optimized operations, and centralized AI collects long-term insights for a higher degree of intelligence and opportunities for automating and simplifying CSP operations and AI-powered automated solutions.
- Network modernization is, first and foremost, the most critical step in providing a competitive customer experience while meeting energy efficiency goals, with AI-native features providing an 11.6% increase in throughput for heavy users and a 15% power savings in MIMO radios.
- Data-driven organizations leverage data and AI early and often, accelerating the digital transformation journey. Here, vendor partnerships can give CSPs a head start in developing cost-effective and best-use oriented approaches.
More data on these insights is available here.