Deep-learning-assisted subwavelength phase imaging based on a quantitative lensless design rule

Conventional phase imaging often faces trade-offs among robustness, measurement speed, and lateral resolution. While diffuser-based lensless systems and deep learning have recently emerged as promising solutions, quantitative characterization of lateral resolution remains limited, and subwavelength half-pitch resolution has yet to be demonstrated, particularly in single-shot configurations. We propose a quantitative design rule that links resolution to diffuser characteristics. The rule is validated through numerical simulations and experimental demonstrations based on deep-learning-assisted frameworks. A subwavelength half-pitch resolution of 342 nm is achieved at a wavelength of 473 nm using the proposed design rule. This framework provides a pathway toward robust, high-speed, and high-resolution lensless phase imaging.