
New research from the University of Surrey has claimed that an “optimal combination of energy-efficient” 5G (mobile broadband) network features, such as AI systems that let mobile mast and antenna base stations go into sleep mode when usage is low and phones that avoid unnecessary background network checks, could cut “indirect carbon emissions” across the UK by c.25 million metric tons.
Regular readers will know that a number of mobile network operators have already deployed at least some of the solutions being talked about in the new research paper. For example, EE have already deployed Cell Sleep technology across their UK 4G (mobile broadband) network (here), while Vodafone have implemented 5G Deep Sleep, 4G Cell Sleep Mode and a Radio Power Efficiency Heatmap (here), among other things.
The new study examines ten energy saving technologies for 5G mobile networks, six that target how base stations operate and four designed to make user devices more energy-efficient. Some of these included AI-driven multi-level sleep modes (this was the most impactful measure), reconfigurable intelligent surfaces (i.e. smart panels that redirect radio waves using little power), cluster-zooming in cell-free MIMO networks (i.e. allowing groups of small antennas to expand or shrink coverage as required and smarter handset signalling etc.
Advertisement
The analysis shows that sectors such as financial services, IT services and computer programming gain some of the largest indirect benefits, reflecting just how much modern industries depend on digital connectivity.
Dr Lirong Liu, Associate Professor in Environment and Sustainability, said:
“Smarter base stations and devices don’t just cut electricity use in telecoms – they reduce indirect emissions in the whole supply chain. The modelling framework allowed us to quantify effects that are usually hidden, especially the indirect emissions linked to electricity use and wider supply chains. It also gave us a clear way to compare different 5G features side by side and identify which combinations deliver the strongest environmental benefits.”
However, the research also suggests that to fully unlock these benefits, 5G policy must “extend beyond coverage and speed targets and encourage the adoption of energy-efficient architectures“. Related measures, say the study authors, could include building energy targets into Ofcom’s spectrum licenses for 5G frequencies, which they say would incentivise for low-power network design and making sure 5G research supports the UK’s broader Net Zero goals. In fairness, the financial savings alone for network operators (i.e. lower energy bills) of adopting such methods are often enough of an encouragement.
Advertisement
Well, yes, things tend not to use power if we power them off.
The problem is that networks are not deploying smart systems everywhere. EE, for example, makes some bands sleep in London during the night, but they come back if usage increases. In more rural areas, they seem to use a dumb system that turns off capacity based on what time is it, not on the needs of the area. Lots of people stuck on the motorway? Someone doing heavy downloads? It doesn’t matter, the bands stay down.
I’m fine with saving power, but the network still needs to work.
One way to reduce emissions would be to let more people use 5G SA. Currently in the UK it is sold as a premium product, but 5G is significantly more power efficient. Allowing devices to use 5G SA as much as possible would cut emissions both at the tower and improve battery life (lowering charging emissions).
Does this really require AI/ML to achieve though? Nearly every cell site is layered these days so when the number of connected devices passing traffic drops below a threshold drop the number of layer on that sector. As the traffic demand grows, wake the layers up again. This can be done in real time just based on traffic requirements, with a pretty simple schedule monitor.
In order not to leave paying customers without the QoS they are paying for it needs to be real-time and not modelled on the past month etc. This really isn’t that hard… Is it?
By all means get AI to write the schedule monitor but you don’t need AI to control it.