This work showcases all-optical computing’s potential to surpass current sensing limitations, offering promising solutions for autonomous driving, high-speed rail communication, radar detection, and satellite navigation.
Wireless sensing and communication have become foundational in both civilian and military applications, with direction-of-arrival (DOA) estimation playing a critical role. Traditional DOA estimation methods, such as the MUSIC algorithm, rely heavily on complex RF circuitry and high-speed digital processing, leading to increased latency, power consumption, and costs. This has driven the search for novel computational paradigms to enhance efficiency and performance in DOA estimation.
Optical computing has emerged as a promising alternative, offering significant advantages in speed, throughput, and energy efficiency. A breakthrough in this field is the development of super-resolution diffractive neural networks (S-DNNs), which allow for all-optical DOA estimation across a wide frequency range. This new approach by team at Tsinghua University, surpasses the Rayleigh diffraction limit in angular resolution, enabling faster and more accurate DOA estimation without traditional RF circuits or digital signal processing.
S-DNNs operate by directly processing spatial electromagnetic waves, using cascaded diffractive modulation layers to achieve precise phase modulation of EM waves. This design allows S-DNNs to determine the angle of a target source with higher accuracy and robustness against noise compared to conventional methods like MUSIC, even requiring only a single snapshot for estimation. Moreover, the researchers have demonstrated that S-DNNs can enhance the sensing capabilities of reconfigurable intelligent surfaces (RIS), a key technology for future 6G communications.
The research also addresses the challenges posed by real-world electromagnetic environments and coherent signal sources. The team developed a broadband training method based on deep learning, enabling S-DNNs to achieve super-resolution DOA estimation across a frequency range of 25 GHz to 30 GHz with an angular resolution of 3°. Experimental results have validated the system’s high accuracy and effectiveness in integrated sensing and communication tasks.
