Self-supervised U-Net-assisted initialization for Gerchberg-Saxton flat-top beam shaping

Gerchberg-Saxton (GS) beam shaping is sensitive to the initial phase, which can slow convergence and degrade reconstruction quality. We present a self-supervised U-Net that predicts an initial phase from a target ffat-top intensity pattern. Without phase labels, the network is trained through an angular-spectrum diffraction model using a composite loss that penalizes intensity nonuniformity, energy leakage outside the target region, and mean-squared error. The predicted phase is then reffned by GS. U-Net provides a common initial-phase predictor for different target shapes without target-speciffc derivation or iterative parameter search, while achieving shaping quality comparable to spherical and geometric initialization. The framework is evaluated for circular, rectangular, elliptical, annular, triangular and pentagonal targets.

Authors

Institutions

Publication Details

Journal
Journal of Optics
Published
2026-11-19
DOI
https://doi.org/10.1088/2040-8986/aec462
Primary Topic
Advanced optical system design
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Self-supervised U-Net-assisted initialization for Gerchberg-Saxton flat-top beam shaping

Xiang Hao, Jiahuan Pan
Journal of Optics
Advanced optical system design
article

Self-supervised U-Net-assisted initialization for Gerchberg-Saxton flat-top beam shaping

Xiang Hao, Jiahuan Pan
article en

Abstract

Gerchberg-Saxton (GS) beam shaping is sensitive to the initial phase, which can slow convergence and degrade reconstruction quality. We present a self-supervised U-Net that predicts an initial phase from a target ffat-top intensity pattern. Without phase labels, the network is trained through an angular-spectrum diffraction model using a composite loss that penalizes intensity nonuniformity, energy leakage outside the target region, and mean-squared error. The predicted phase is then reffned by GS. U-Net provides a common initial-phase predictor for different target shapes without target-speciffc derivation or iterative parameter search, while achieving shaping quality comparable to spherical and geometric initialization. The framework is evaluated for circular, rectangular, elliptical, annular, triangular and pentagonal targets.

Journal of Optics
Zhejiang University (CN)
Advanced optical system design
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon