Learn how machine learning is making a dent in 6G, Wi-Fi, networking, chip design and more.
Artificial intelligence (AI) and machine learning (ML) are seeping their way into all corners of industry, and wireless communication is no exception. Engineers are beginning to see the impact in everything from networking to chip design. And as the industry moves towards 6G, AI is only going to become more prominent.
Companies from Cisco to Keysight Technologies to Qualcomm are all exploring the use of AI in communications systems: troubleshooting, power saving, channel estimation, MIMO detection—the list goes on. And as much as AI is already changing wireless design and providing helpful tools to engineers, there is even greater potential for the technology in the future.
Streamlining wireless systems
The first applications of AI in the wireless industry have been for streamlining existing systems.
“The industry is using AI techniques to improve upon existing wireless communication, engineering of systems and networks,” Houman Zarrinkoub, principal product manager at software company MathWorks, told engineering.com.
One of the ways this is often seen is within network design space. Artificial intelligence is helping designers manage larger networks and systems that can be cumbersome to handle.
“Resource allocation, scheduling and dividing finite resources within a large subscriber base is one of those applications where AI is shining right now,” Zarrinkoub says.
Wi-Fi is among the wireless standards that is starting to reap the benefits of AI. “There’s a new standard called 802.11az which is dedicated toward positioning and localization, and there is a new standard called 802.11bf which is essentially designed for wireless sensing. Both of these things have AI algorithms associated with them,” Zarrinkoub says.
AI is also impacting wireless chip design, part of a broader industry trend. In 2020 Google developed a reinforcement-learning algorithm that designed chip floor plans for Google’s TPU. Google Deepmind announced further developments this year, using AI to design specialty semiconductors. Numerous organizations are taking on the task of pushing AI-powered chip design even further.
For wireless chip design, AI is helping at the power amplification step of signal transmission. As society has moved through the generations of wireless from 3G to 4G to 5G, the transmission bandwidth requirements have gone up. This is where Zarrinkoub sees AI-application potential, especially as this requirement continues to go up for 6G.
“Imagine you have one gigahertz bandwidth. These systems are physical systems. Maintaining linearity, meaning that energy is distributed evenly at the first frequency until the last frequency, is a nightmare,” Zarrinkoub says. “But nowadays everybody we talk to on the device side is asking for the application of AI in digital predistortion and in maintaining linearity of a power amplifier.”
AI design tools
Companies like MathWorks, DeepSig and Ekahau are providing tools to support the next era of AI-powered wireless design. MathWorks’ Matlab software offers machine learning, statistics, deep learning and reinforcement learning toolboxes.
“In the Matlab environment, we provide all the algorithms that are out there. We grow them based on the new innovations that are helping with either training-data-driven approaches or simulation-driven approaches. Then we use those foundational AI applications and products to apply it to the vertical applications,” Zarrinkoub says.
The company website touts that these applications could help generate training data in the form of synthetic and over-the-air signals, label signals collected from wireless systems, or augment signal space by adding RF impairments and channel models to user-generated signals.
One of the common concerns with machine learning algorithms is the concept of “black box” AI. This is an algorithm that can operate and make choices, but cannot show how those outcomes were reached. Zarrinkoub says it has been a focus in Matlab to make it accessible and clear where AI-based results are coming from.
“If you are in doubt of what we are doing underneath the hood, you open the hood and you see what algorithm was used for training, testing, measurement and so on,” Zarrinkoub says.
Future impacts of AI on wireless
The 3rd Generation Partnership Project (3GPP), the standards body behind 5G, is currently studying the involvement of AI in the communications industry. The group unites seven telecommunications standard development organizations to create technical specifications and technical reports for mobile systems. Although not creating AI and ML models themselves, the frameworks and evaluation strategies made by 3GPP have the potential to guide and propel the future use of these technologies in the industry. Their newest standards release is expected soon.
“They are some promising applications of AI in there,” Zarrinkoub said.
Many future applications of AI and ML are focused on optimization. They prioritize saving designers time and reducing time for decision making in systems. For example, although classical techniques are still being used in the coordination between the base station and user equipment, AI is being discussed as a possible way to improve the system.
“With AI the amount of information sent between back and forth is going to be much reduced. Essentially, that process is going to be a lot cheaper, less computationally intensive and much cheaper in terms of spectrum,” Zarrinkoub said.
While there is focus on bringing AI solutions into the mix as soon as possible, many organizations are also concentrating on potential larger scale adoption in the 5G to 6G transition. As the industry shifts to 6G, everyone is hoping to bring in as many AI- and ML- driven design optimizations as possible.
“As AI models and testing best practices mature, there is no doubt that AI will revolutionize wireless communications in the next 5-10 years,” Sarah LaSelva, Keysight Technology’s director of 6G marketing, said in a company blog earlier this year.
No matter the speed of adoption, AI has definitely found a home in streamlining the wireless communications industry. As Zarrinkoub says, this is because many of the issues that engineers face are a perfect match for the potential of AI-powered technologies.
“In the wireless domain, the problems that are inherently nonlinear or inherently multivariate, where the number of parameters is substantially bigger than we are used to, these are the problems for which AI provides all the right solutions,” Zarrinkoub says.
