Simple-Diffraction-Based Deep Learning to Reconstruct a High-Dimensional Orbital-Angular-Momentum Spectrum Via Single-Shot Measurement
Haoxu Guo, Xiaodong Qiu, Lixiang Chen
Abstract
Orbital-angular-momentum (OAM) complex-valued spectrum reconstruction routinely necessitates the measurements of a complete set of observables, which is rather time consuming and inefficient for high-dimensional OAM superposition state. Here we break this barrier to realize the full reconstruction of a high-dimensional OAM complex-valued spectrum via only a single-shot measurement, which benefits from the use of a residual convolutional neural network for learning the diffraction behaviors. In our experiment, the high-dimensional OAM states are diffracted by a simple aperture of pentagram that breaks the OAM conjugate symmetry. We achieve fidelity over 97.8% and 92.1% for high-dimensional pure and mixed OAM states (up to $d=15$), respectively. We also confirm the excellent robustness under strong noise environment. Our method will find potential applications in high-dimensional information encoding, such as large-alphabet communications, quantum imaging, and real-time quantum sensing.