Can 6G Learn the Lessons of IoT?

May 3, 2023

Written by Alex Lawrence

“Coverage is the most important thing for future networks,” commented Dritan Kaleshi, of the UK’s Digital Catapult. “For a long time I thought otherwise – that it was speed or capacity – but ultimately coverage makes so many services possible.”

Kaleshi was speaking at last week’s 6GSymposium in the UK. He was not alone in promoting the idea that being able to deliver truly ubiquitous coverage might be the ‘superpower’ for 6G.

This is just one area where the Internet of Things [IoT] has a head-start. 6GWorld spoke to Dima Feldman, VP Product Management and Marketing at Sony Semiconductor Israel – formerly Altair Semiconductor – about the work they’re doing in providing chipsets for IoT. The overlap with a range of 6G preoccupations is striking.

“Really the guys who deploy IoT devices should not care about the standard,” Feldman commented. “They should care about their business case, how they monitor water or energy consumption, how they optimise that, how they detect leakage in our water pipes.”

Perhaps unfortunately, this is in fact the case.

“The biggest market we have today is utilities. Every house has two or three meters so our numbers are substantial, but there is no unified standard. Everybody uses different technology. Some use Wi-Fi, some use Wireless M-Bus; many more use cellular, some use LORA and some use other proprietary stuff,” Feldman explained.

This context means that the chipsets being developed have to have the capability to attach to multiple technologies, whether cellular, satellite or dedicated IoT.

“We added additional technology like device-to-device communication,” Feldman explained.

The use case is fairly clear; to make sure that IoT devices are able to connect regardless of the environment they find themselves in.

“For example, if your platform’s cellular but in some locations there is no cell coverage, you have a fallback solution. Or if you have your main deployment in mesh, one of the problems you is you don’t have the coverage you get with cellular; so 5% of the devices go to the cellular level.”

While there have been discussions about creating a new architecture for a network of networks across fixed, mobile and satellite, the fact that we already have devices selecting the best-available network demonstrates that there’s another way forward. To scale up from low-data, low-power devices to ones that demand more of the network may require some support, but that could be more of a focus for operational and commercial systems to enable ad-hoc attachments across different kinds of network.  

Inbuilt Intelligence

“AI’s a range, right? There are very different processes and AI capabilities,” Feldman began. His team are working on AI loaded onto their chipsets; given the focus on AI in evolving telecoms networks it was a natural area of questioning.

Unsurprisingly, the difference lies in the cost and power consumption of chipsets. At the low end, Feldman identified a few use cases based on sensor processing.

“It could be microphone data for detecting noise from an air conditioner or having an AI processing accelerometer data. So having one- or two-dimensional data is very, very low-power, but you can do important things like detecting that your AC or your car is making strange noises.”

At the same time, the training of such AIs about ‘normality’ does require a larger initial power consumption. It’s far from the huge amount of energy and data required for more general AI, however.

“You train it on the edge, you create a number of samples. It’s totally different way of engineering – fewer formulas and more data sets,” Feldman commented. It really simplifies the way you deploy the solution.”

Significantly for the telecoms industry at large, being able to offload large numbers of simple processes to the device means that it reduces network traffic; rather than having a permanent feed, only the unusual input data need be sent and flagged for attention or further processing. Because of the power demands of radios, this is a useful way to operate low-power devices.

On the more advanced end of the spectrum, Feldman pointed to visual image processing as a popular use case. Once again, the AI on the device is being used to decide what images or video footage need to be sent to a cloud-based AI for interpretation or reaction. He gave an example in a retail environment, where somebody has picked up a bunch of bananas.

“What if they put the bananas back? Whoa, maybe something is wrong with the bananas on the shelf. Somebody should look at that.”

This is useful information to trigger a check in-store. However, the security camera market alone is huge and booming. According to Mordor Research, the sector’s value will exceed $21bn by 2026.

“There’s absolutely no way you can transmit a feed from millions of millions of cameras to the cloud. There’s not enough bandwidth – not in 4G, not 5G, not 6G or 7G,” Feldman asserted. “You have to do this processing on the edge, inside the camera.”

This mindset is similar to that driving the development of semantic communications, insofar as the focus of telecoms shifts from sending data to sending information, in order to improve the efficiency of communication. Under current paradigms, any data packets lost or faulty need to be retransmitted.

This is far from a human way of communication, where we can make out messages with high levels of certainty even with some interference or muffling, especially given an understanding of the context of the message. By applying intelligence through the network, traffic can be reduced significantly.

Until that is ready for commercial usage and mainstream, however, minimising the quantity of messages sent is a crucial step forward in order to deliver effective services without over-burdening the network.

Sustainability & Power Consumption

While there is pressure on vendors and operators today to reduce power consumption – both because of the environmental impact and the cost – this too is an area which has been pioneered in the IoT for sensors which can last for years on one battery or even harvest ambient energy.

The inclusion of AI into such sensors is liable to be valuable, for the reasons outlined above, but it also means minimising the amount of energy used to fit within the available energy budget.   

“Some of the devices have to last for 15-20 years,” Feldman pointed out.

“Size and power consumption are critical. You may want to have a camera which is solar powered – with this size of solar panel and this type of accumulator – and all the processing has to fit within that budget.”

While the energy efficiency of an individual sensor or IoT device may have an impact only on the business case for that specific device, the cumulative impact can be significant. 2022 Research by Clarion Security Systems suggests that, for example, in London alone there are over 940,000 CCTV cameras.

“So you do something within one watt rather than 10 watts; that translates to the megawatts of power over the year.”

Governments have been setting out their visions for a high-coverage, low-energy future for their connected societies. It’s ironic that techniques developed for the humble IoT sensor might point the way forwards.

Drop us a line

Interested in contributing to 6GWorld.com? Let's connect to learn more about your ideas.

Drop us a line

Interested in contributing to 6GWorld.com? Let's connect to learn more about your ideas.

Recent Posts

NVIDIA Launches 6G Research Platform for Cloud

NVIDIA Launches 6G Research Platform for Cloud

NVIDIA announced this week a new 6G research platform that, according to the company, will empower researchers with a "novel approach to develop the next phase of wireless technology." Ansys, Arm, ETH Zurich, Fujitsu, Keysight, Nokia, Northeastern University, Rohde...

Pin It on Pinterest

Share This