Reducing RAN Energy Usage with Artificial Intelligence

Energy prices are rising, and mobile network operators are feeling the pinch. In July 2022, Vodafone indicated it may have to raise prices to cope with rising electricity costs; Telenor said in September that increased power prices were damaging its profitability for the year; and SoftBank in Japan followed suit in November.

In the face of these energy price increases, as well as pressures to reduce carbon emissions, operators need to drastically lower their power consumption — this means understanding where and how energy is being used. According to the market intelligence unit of the GSMA, 73% of energy consumed by operators is used in the radio access network (RAN) to power radio transmissions, with radios providing signal coverage often over many kilometres and each radio site supporting many thousands of potential users, devices and data sessions. The rest is consumed by core and transport networking and data centre operations. Clearly, the RAN needs to be the primary focus of energy saving.

This comes at a time when operators around the world are focussed on 5G deployment, and this is an escalating challenge for mobile networks. Life cycle data shows that every time a new generation of technology is introduced, there’s been a corresponding increase in energy consumption. Roll-outs of 5G drive energy usage because of the greater antenna power and capacity requirements of 5G radios, which are deployed in increasingly dense network coverage patterns to deliver the capacity needed to support ever-growing data traffic. This results in energy usage going up, not down. In fact, projections by the GSMA suggest that from 2020 to 2030, power consumption by mobile networks will triple if RAN energy usage remains unchecked.

With the disproportionate amount of network energy usage in the last mile, reductions in the RAN could have the biggest immediate impact. This is because the RAN in 4G or 5G networks is typically configured to be “always on” — fully powered for uptime and resilience obligations to fulfil service level agreements, designed to support the highest possible data throughput. So, the question is: could this be delivered with greater visibility in traffic monitoring to ensure there are no dropped packets or frames without compromising signalling or radio availability?

We think the answer is integrating artificial intelligence (AI) into the mobile network. There are already some innovative approaches to reducing RAN power consumption. ZTE, for example, has developed its AI-based PowerPilot Pro solution, which has been nominated for best mobile innovation for climate action at the Global Mobile Awards, also known as the GLOMOs, handed out by the GSMA at its annual MWC event in late February for industry achievements.

ZTE’s solution uses network telemetry and automated decision-making to dynamically scale radio resources, based on current demand from network users, enabling operators to make energy efficiency adjustments in near-real time and meet performance requirements. Also, it’s possible to take advantage of fluctuations in network traffic to dramatically improve power efficiency. Rather than just examine data points from a single domain, PowerPilot Pro analyses data across domains to recognize energy-saving opportunities and predict network traffic trends. It also performs multicell coordination to determine optimal energy-saving policies.

This is a good example of how using AI in the RAN can help operators analyse traffic patterns to predict opportunities for further energy savings in periods of lower demand. The fast reaction time of the solution means that idle resources enter a sleep mode when customer traffic is low and resume normal operation when traffic increases.

Powering down a base station at night isn’t easy. In the RAN, the active antenna unit is a highly integrated, complex and power-hungry item of equipment. In terms of power consumption, its chipsets don’t fluctuate much with variation in traffic load, so even when there’s no traffic its minimum power consumption without activating any power-saving function is still about half the peak power level.

At quiet times for traffic, like between 2 AM and 4 AM, radio heads can be put into a deep sleep mode, shutting down chipsets and power amplifiers to reduce power to about 10% of peak load. The fronthaul uses the enhanced Common Public Radio Interface (eCPRI) to connect the active antenna unit with the baseband unit where its radio signals are processed. This connection is normally kept awake so the active antenna unit remains visible to the RAN management system and can be woken from its deep sleep. But by using a hibernation technique for the antenna, the eCPRI module can be turned off, reducing energy consumption further to less than 1% of that in basic deep sleep mode.

ZTE’s approach is enabled by a cloud-based platform and involves extending AI to each base station and creating large-scale AI training to offer a responsive system that allows for action in near-real time. The result is RAN-wide AI analysis in a centralized cloud with data flows delivered in milliseconds — an intuitive system that learns from past activity to be more predictive of future needs.

This intent-based autonomous control of mobile networks is a central part of creating greener networks and carries forward into ZTE’s vision for 5G in other sectors. It believes 5G has the potential to be a major driver of sustainability — particularly when used with other emerging technologies such as AI, automation, data analytics and edge computing. Adoption of 5G can help accelerate digital transformation in enterprises and specialized industries, supporting innovative applications, operational efficiencies and energy savings.

And this is critical in proving that, even with rising data traffic on mobile networks, 5G energy usage will be more than balanced out by the net carbon emissions savings it will deliver, helping other industries reach their net zero targets. For example, more dynamic allocation of resources in an organization could mean supporting more efficient supply chains enabled by predictive analytics or automated, intelligent vehicle management in all manner of smart traffic, logistics and transportation environments, reducing the use of fossil fuels. Or it could be used to enable data processes that decrease the need for operational or leisure travel, such as remote medical diagnosis, drone-based site surveying or virtual training or tourism.

Although 5G can be the foundation for a digital economy, it must also be a green economy. The IT industry must further lower energy consumption in transmitting, processing and storing ever-rising volumes of information by providing infrastructure that’s more intelligent, efficient and carbon-aware, as well as delivering an ever-increasing quality of service.

With the use of AI, data analytics and other technologies, operators can get a deeper understanding of when power is and isn’t needed, and then use that contextual information to implement differentiated energy-saving policies. Approaches like ZTE’s PowerPilot Pro can help reduce RAN power consumption without compromising performance, helping mobile operators cut carbon emissions as part of larger corporate sustainability strategies.