Machine learning based accurate recognition of fractional optical vortex modes in atmospheric environment

Abstract

Optical vortex beam with fractional orbital angular momentum (OAM) has great potential to increase the capacity of optical communication and information processing in classical and quantum regimes. However, atmospheric turbulence (AT) in free space distorts the helical phase-front of vortex beams and causes the mode diffusion, seriously hindering the practical application. Herein, using a convolutional neural network approach with an improved residual neural network architecture, we overcome the hurdle to give the accurate recognition of the fractional OAM in the AT. As demonstrated on the petal interference patterns, a type of hybrid beams carrying double OAM modes is utilized to provide two controllable degrees of freedom for greater recognition of more subtle OAM modes, e.g., the fractional topological charge number l and the angular ratio n. Our studies show that with various l and n, the recognition accuracy of OAM over 20 000 images is as high as 85.30% even under the strong AT parameter (Cn2 = 5 × 10−14 m−2/3) and the long propagation distance (z = 1500 m). Our findings represent a remarkable achievement toward highly accurate recognition of fractional OAM with broad bandwidth in the atmospheric environment, expanding the applications for the general interest of machine learning based OAM optical communication.

Deep Learning Recognition of Orbital Angular Momentum Modes Over Atmospheric Turbulence Channels Assisted by Vortex Phase Modulation

“…8, where our results decline much slower as the structure constant or the propagation distance increases. For fraction recognition, the recognition accuracy under z = 1.5km and C 2 n = 5 × 10 −14 m −2/3 in [30] is 85.30%, while our result under z = 1.6km and C 2 n = 5 × 10 −14 m −2/3 is as high as 99.72%.…”

“…To demonstrate the performance enhancement by the vortex phase modulation, we compare the recognition results in this work with those from [28] and [30]. For integer recognition, performance comparisons between our work and [28] are presented in Fig.…”

“…Therefore, we decide to use a twelve-layer CNN in this work. It provides better recognition performance than the CNNs with fewer layers did in [24], [28], [42] and requires fewer computational resources than the CNNs with more layers did in [30], [43].…”

“…Based on the well-known DenseNet, this OAM-recognition network exhibited high resolution of the OAM state interval (i.e., the step length of topological charges) as 0.01. The recognition of fractional OAM modes in turbulent environments at the resolution of 0.1 was reported in [30], which is realized by a ResNet-based CNN model. Apart from recognizing OAM modes in fractions other than integers, distinguishing between positive and negative topological charges is also of interest to the researchers.…”

Adaptive demodulation by deep-learning-based identification of fractional orbital angular momentum modes with structural distortion due to atmospheric turbulence

“…The author improved anti-turbulence ability by combining a method of diffraction preprocessing of a two-dimensional fork grating and achieved the recognition accuracy of 99.1% for the mode interval of 0.1. In addition, deep-learning-based detection of hybrid beams carrying fractional topological charge and the fractional angular ratio was investigated, which showed accurate recognition of fractional OAM with broad bandwidth in atmospheric environments 28 .…”