Assembling Reconfigurable Intelligent Metasurfaces With a Synthetic Neural Network

Reconfigurable metasurfaces have provided us with extraordinary ways to manipulate electromagnetic waves at subwavelength scales and taught us how to implement various tasks in a wide range. As its next generation, intelligent metasurfaces have recently attracted much attention as they are endowed w...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on antennas and propagation 2024-06, Vol.72 (6), p.5252-5260
Main Authors: Lin, Pujing, Qian, Chao, Jia, Yuetian, Wang, Zhedong, Wu, Bei, Huangfu, Jiangtao, Chen, Hongsheng
Format: Article
Language:English
Subjects:
Algorithms
Assembling
Changing environments
Convolution
Deep learning
Electromagnetic radiation
Fixed objects
Intelligent metasurfaces
Inverse design
Kernel
Machine learning
Metasurfaces
Neural networks
Reconfiguration
synthetic neural network
Task analysis
Training
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Reconfigurable metasurfaces have provided us with extraordinary ways to manipulate electromagnetic waves at subwavelength scales and taught us how to implement various tasks in a wide range. As its next generation, intelligent metasurfaces have recently attracted much attention as they are endowed with intelligence to self-adapt to ever-changing environments. So far, although there are a number of works about the creation of deep-learning algorithms to drive intelligent metasurfaces, they are mostly about fixed objects and thus constrained by specific applications. In this work, we propose a synthetic neural network to break the limit that conventional inverse design algorithms only work for certain metasurface configurations. The dynamic assembly of reconfigurable metasurfaces in physical space is directly mapped to the network recombination in algorithm space. This is particularly useful because it can greatly save training data and avoid repeated training of the network. We apply it in the inverse design of metasurfaces with a relative error of less than 13% and experimentally demonstrate reconfigurable microwave metasurfaces with flexible deployment. Our work opens up a promising avenue for overcoming the application-bound limitations of conventional inverse designs, particularly in the dynamic and adaptive context of intelligent metasurfaces.
ISSN:0018-926X
1558-2221
DOI:10.1109/TAP.2024.3395909