Using AI to make 5G more sustainable

As the digital renaissance starts to unfold, it is connectivity such as 5G that provides a foundation to almost every emerging digital technology.

Smart sensors, robotics, virtual reality, artificial intelligence, to name a few, all depend on the low latency of 5G to uphold a better quality experience with limited lag.

However, energy consumption of cellular networks is concerning. The telecoms industry uses around 3% of the energy consumed globally according to the GSMA. Comparably, the aviation industry uses about 2%.

With digital technologies growing, cellular energy consumption is only expected to increase due to the need for more infrastructure needed – a concern that smart devices and ICT infrastructure firm Huawei plans to handle before digital technologies truly boom.

“To tackle this problem, we need to look into the part of the network that consumes the most energy, and this is the radio access network (RAN)”, explains Nicola Piovesan, senior researcher at Huawei Technologies in France.

RAN-away consumption

RAN is the part of a mobile network that connects end-user devices such as smart phones to the rest of the network through radio waves.

This element alone makes up for 73% of the energy consumed in telecoms, while 13% is in the core of telecoms, 9% in data centres, and 5% in the rest.

To break this down further, most of the energy consumed within the RAN is in the base station – “a base station typically is comprised of a basement unit that is responsible of the digital signal processing and the transmission of the information in a physical way.”

The issue with base stations is that most (54%) are powered by carbon-emitting sources, according to a GSMA study.

43% are powered by the traditional grid, and in developing regions where renewables and grid is less prevalent, the base stations are powered by diesel (making up 11%).

“There is a big push from mobile network operators today to achieve net zero emissions in the next ten to 20 years and this typically is in agreement with electricity providers,” says Piovesan.

So, while network providers are increasingly working to power base stations with renewable energy, it is also possible to look at usage of base stations and limiting powering when they are not actively handling data and signalling.

Traditional mobile networks such as 4G spend only about 15% to 20% transferring data, and the rest is wasted because of heat loss in power amplifiers, the equipment running when no data is being transmitted and inefficient cooling systems and battery units.

Although 5G networks are around four times more energy-efficient than 4G networks, the total energy consumption is about three times larger as a result of 5G’s demand.

“The energy consumption could be cut down by half, according to a McKinsey study if networks introduced smart shutdown techniques, and procuring green energy,” explains Piovesan.

Currently, Huawei is using a basic machine learning model based on a general base station for its smart shutdown techniques. The issue with this is that base stations can be dramatically different in their use depending on the population around it and so ideal shutdown times technically vary.

So, by harmonising the data collected by machine learning and using it with AI, together with IoT based-monitoring to support smart energy management, the team created a model that can better accurately predict the switch-off of machinery, and city infrastructure and thus use less energy.

“Due to its fundamental nature, the [AI model] can be adopted to model other types of [base stations] deployed in different multi-vendor networks,” writes Piovesan in a paper proving the use of their AI model.

The combination of AI, machine learning, cloud, and IoT devices, as well as the 5G itself, can help businesses themselves underpin their own private 5G networks in factories, mines, and other environments to make them more sustainable.

As IoT sensors are connected to digitally-connected machinery, once that machine is not in use the 5G connectivity to that device can be turned off and switched to another.

“It allocates precisely the network and computing resources that are required for a certain industry and a certain task, for as long as they are required,” Huawei explains in its paper. “Once the task is over, the virtual slice of resources can be switched off and be re-used somewhere else.”

Challenge

At the AI for Good Global Summit last month, the UN’s telecoms agency, the International Telecommunication Union (ITU), and Huawei announced its 5G energy modelling challenge in the hope to develop machine learning models to reduce the energy consumption of base stations for any manufacturer.

Closing at the end of September (30^th), the challenge hopes to “raise awareness of knotty problems in ICT, encourage researchers and experts worldwide to contribute to the maturity of applicable solutions, and present how AI capabilities can be leveraged to deal with real-life issues,” explains Xi Zheng, researcher at Huawei’s Global Technical Service Department and coordinator of the challenge.

Winners will receive cash prizes as well as collaboration opportunities with Huawei.

Seizo Onoe, director of ITU’s Telecommunication Standardization Bureau, said it is important to observe “how ICT and AI solutions, when combined, can provide effective solutions for concrete action, at a time when tangible progress is urgently called for on a global scale”.