How will AI drive 5G RAN energy efficiency?

Learning from the past to predict the future and optimize the present—AI can match RAN resources with customer demand and, in turn, optimize for energy efficiency

At roughly the midpoint of the 5G cycle, communications service providers (CSPs)  are navigating a macro environment marked by the need to more effectively monetize existing network-enabled capabilities while making durable, futureproof investments that set them up for long-term success. As 5G ROI continues to come into focus, the imperative is to pull any and all available levers that serve to reduce operational costs. While it could potentially complicate these concurrent transformations, the rise of artificial intelligence (AI), if employed strategically and holistically, could indeed help align these short- and long-term goals with the knock-on effect of helping CSPs quickly improve network energy efficiency in support of Net Zero initiatives. 

Embedding AI throughout the RAN will come with a long tail marked by streamlined network operations that enhance performance while driving down costs. An early use case gaining traction is around using AI to dynamically turn off radio components when they’re not needed to meet demand, thereby reducing energy consumption and attendant costs without impacting user experience. And for good reason. According to GSMA Intelligence, the majority (87%) of an operator’s energy consumption comes from the RAN, so it’s an obvious place to start. 

Some examples of how AI is being used to affect significant energy savings: 

• Umniah Jordan deployed Ericsson’s Intelligent RAN Power Saving solution and demonstrated 20% daily power saving capabilities in its 5G network. 
• Three UK is using Ericsson AI-powered hardware and software, including the dual-band Radio 4490 and software features focused on power-saving, to improve energy efficiency up to 70% at select sites. 
• Taiwan’s Far EasTone trialed Ericsson’s Service Continuity AI App suite to cut power consumption without impacting performance, and saw a 25% savings in daily RAN energy consumption. 

There’s an important distinction between the generative AI solutions taking up mindshare, and necessitating investment in power-hungry GPU clusters, and the approach to AI and ML being used in radio access networks. In the latter case, the general trend is that any increase in energy needed to implement AI is more than offset by the decrease in energy consumption AI delivers. 

Regardless, the World Economic Forum rightly points out that AI-related energy consumption is accelerating by double digits and, as such, recommends that sustainability stays top of mind as AI evolves. Beena Ammanath, a board member for WEF’s Center for Trustworthy for Technology, put it this way in a blog post: “As we stand at the intersection of technological innovation and environmental responsibility, the path forward is clear. It calls for a collective endeavor to embrace and drive the integration of sustainability into the heart of AI development. The future of our planet hinges on this pivotal alignment. We must act decisively and collaboratively.”

To Ammanath’s point about the importance of collaboration, and grounding that in the world of telecoms, Ericsson is a founding member of the AI RAN Alliance, alongside firms including Microsoft, NVIDIA, SoftBank, T-Mobile and others. A pillar of the group’s mission “evolving networked sustainability…to make our existing systems more intelligent, efficient and reliable, empowering us to expand the meaning of sustainability in multiple dimensions.” 

And this is just the beginning

To understand the evolutionary outlook for AI in the radio access network (RAN), RCR Wireless News spoke to Ericsson Director of AI and ML Strategy Ayodele Damola. First things first, he said the overarching strategic goal is to “ensure we leverage this technology to provide the best capabilities for our customers.” He sketched out a four-step AI journey: 

• Rules-based automation—responses to all situations have to be taken into account when designing the automation
• Autonomous features—AI/ML adapts to unprecedented situations
• Cognitive intent-driven networking—automation agents are instructed on what to achieve and left to figure out how to achieve it
• And zero-touch networking—AI-native automation governs automated network lifecycle management 

As Ericsson continues on this journey toward zero-touch alongside its customers, Damola pinned the current state as between the presence of autonomous features and cognitive intent-driven networking. “From an operator’s perspective, the operator will communicate a high-level ambition or desire…An example of an intent is, I want my fixed wireless access subscribers to have 10 Mbps throughput and latency of no more than 5 milliseconds. The AI functions embedded in the network will perform the required actuation steps to meet these requirements.” He also noted that an operator could also communicate a specific parameter instructing the system to deliver a particular service with particular KPIs “with the minimum energy consumption.” 

Vodafone uses AI to reduce energy consumption across a MIMO cluster

Back to the present—operators around the world are working to align goals around network modernization, monetization and sustainability through holistic technology investment strategies, including the use of AI for various network optimizations. In one example Damola laid out, Vodafone used AI-powered sleep mode across a radio cluster. AI algorithms applied to the baseband to forecast traffic patterns compared projected traffic utilization with actual physical resource blocks (PRBs), and made a decision whether to turn down antenna branches, power amplifiers and other electrical components. 

With no manual configurations necessary, Vodafone saw a 14% savings in energy consumption per site, which outperformed the benchmark of manual configuration. The operator’s KPIs were maintained throughout. Vodafone Group Head of Radio Products Paco Martin said, “The opportunity for machine learning in this case is the ability to identify patterns. It’s a fantastic way to learn from experience.” 

Damola pointed out that initially the savings in energy consumption was around 9%, but, “The system does learn over time based on what it sees,” meaning more data and more learning resulted in further improvements to the outcome. 

With the ability to learn from the past to predict the future and apply that knowledge to optimize the present, Damola said the underlying model really “forecasts the traffic utilization one minute in the future. At any given point in time, the system makes a prediction for the next one minute.” 

As the technology continues to bear fruit in the field, and deliver improved outcomes over time, the focus shifts to how organizations adapt to a new, automated way of working; and how potentially siloed business or functional groups align their own objectives to include energy efficiency. To this point, Damola characterized AI as an opportunity for operators to adopt a holistic view of organizational directives, including alignment of energy efficiency, NetZero and network performance-related measures. Seeing in the real world that AI can deliver increased energy efficiency without compromising user experience will gradually address the “question of trust,” Damola said. 

After the question of trust, he suggested perhaps it’s a question of incentives. “The picture might change if everyone in the company is incentivized based on energy efficiency.” But, regardless, as the world continues to warm and net-zero deadlines near, really, “It’s a question of time,” he said.