Scientists at China’s Southeast University have developed an innovative method to use deep learning to control electromagnetic waves faster and easier. The advancement focuses on programmable metasurfaces—ultrathin materials designed to manipulate waves such as light and radio waves. The research results led by Professor Cui Tiejun were published in iScience.
Programmable metasurfaces are known for their ability to shape electromagnetic waves, but designing patterns that control electromagnetic waves has been a slow and challenging task. Electromagnetic waves are forms of energy, such as light or radio signals, that travel through space. “Our model can calculate these patterns almost instantly by simply specifying what the waves should look like,” explains Professor Cui. Their approach combines cutting-edge deep learning techniques—a type of artificial intelligence that teaches computers to recognize patterns and make decisions—with physical models of the behavior of electromagnetic waves. This approach ensures that the results are both accurate and practical. The researchers demonstrated its success in tests, showing that it works for both simple and complex wave patterns.
Metasurfaces are essentially ultrathin materials designed to interact with electromagnetic waves in precise ways. Unlike traditional materials, they can be programmed to perform specific tasks, such as focusing signals or filtering out unwanted frequencies. Previously, these surfaces could not be adjusted after fabrication, limiting their usefulness. The invention of programmable metasurfaces allows engineers to make these adjustments electronically, enabling a wider range of applications. However, figuring out how to create the optimal waveform requires a complex, time-consuming process that often involves an iterative approach to gradually refining the solution. These methods are impractical for real-world applications. New approaches address this problem by using deep learning to bypass these traditional challenges.
The new system has several benefits. It works almost instantaneously, which means it can respond quickly to changing needs. By incorporating physical principles into deep learning models, it also does not require the vast amounts of labeled training data or simulations required by older methods. Training data refers to examples that an AI system learns to make accurate predictions. The process uses deep learning systems to calculate the arrangement of surface components and simplified physical models to predict the behavior of waves. Tests have shown that the system can produce patterns in an instant, a significant improvement over traditional techniques that can take hours.
To see how well their method worked, the researchers compared it to an older method called dyad optimization, which was inspired by the collective behavior of animals such as birds or fish as they search for food. Computing technology. They found that their deep learning model not only ran faster, but also created more efficient wave patterns. By eliminating unnecessary data preparation and using faster processes, this approach is more practical and powerful than earlier solutions.
“Our experimental results show that this method can reliably design waveforms for real-time applications, such as scanning objects or improving wireless communications,” said Professor Cui.
The technology has the potential for a wide range of uses, including smart sensors, devices that collect and respond to environmental data, tracking systems and other applications that require rapid adjustments to electromagnetic waves. The team also showed how it could be used in scenarios that require constant updates, such as tracking moving targets with focused energy beams.
Despite their success, the researchers admit there are still areas for improvement. For example, their model assumes that changes in surface composition only affect the phase or timing of waves, without accounting for more complex interactions. This simplification works well, but may overlook some details that could improve accuracy. Researchers are exploring ways to improve the system for better performance.
This breakthrough represents a major advance in the technology for controlling electromagnetic waves. By making this process faster, easier, and more adaptable, this research opens the door to new applications in communications, sensing, and more.
Journal reference
Bao Jianghan, Li Weihan, Huang Siqi, Yu Wenming, Liu Che, Cui Tiejun. “Physics-driven unsupervised deep learning networks for programmable metasurface-based beamforming.” iScience, 2024.
About the author
Cui Tiejun (IEEE Fellow) earned bachelor’s, master’s, and doctoral degrees. He received his PhD degree from Xi’an University of Electronic Science and Technology of China in 1987, 1990 and 1993 respectively. He joined the Department of Electromagnetic Engineering of Xi’an University of Electronic Science and Technology in March 1993 and was promoted to associate professor in November 1993. In July 1997, he joined the Computational Electromagnetics Center of the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign, first as a postdoctoral researcher and later as a research scientist. In September 2001, he was appointed Changjiang Professor in the Department of Radio Engineering, Southeast University. In January 2018, he served as chief professor of Southeast University. He is an academician of the Chinese Academy of Sciences. He is the author of Metamaterials: Theory, Design, and Applications (Springer, November 2009), Metamaterials: Beyond Crystals, Amorphous and Quasicrystals (CRC Press, March 2016), and Informative Metamaterials ( Cambridge University Press, 2021) and other books. He has authored or co-authored more than 600 peer-reviewed journal articles, with more than 62,000 citations (H-factor 122), and has been granted more than 150 patents. His research was selected as one of the most exciting peer-reviewed optical research in 2016 by Optics and Photonics News magazine, one of the top ten scientific breakthroughs in China in 2010, and was selected as a “research highlight” by a series of journals. His work has been widely reported in Natural News, MIT Technology Review, Scientific American, Discover, and New Scientist. In 1995, he received a research scholarship from the Alexander von Humboldt Foundation in Bonn, Germany, and in 1999, he received the Young Scientist Award from the International Union of Radio Sciences.
Browse the car Earn a bachelor’s degree in engineering. Bachelor’s and Ph.D. degrees in Information Science and Technology. degree from Southeast University, Nanjing, China, in 2015 and 2022, respectively. Currently, he is a Zhishan postdoctoral fellow at Southeast University. Selected into the 2024 list of the world’s top 2% scientists (Internet and Telecommunications) released by Elsevier Publishing Group. His research interests include computational electromagnetics, metamaterials, and deep learning. Committed to using artificial intelligence technology to solve electromagnetic problems, including ISAR imaging, holographic imaging, inverse scattering imaging, automatic antenna design, diffraction neural networks, etc.
