Deep-learning-based high-resolution recognition of fractional-spatial-mode-encoded data for free-space optical communications

Abstract

Structured light with spatial degrees of freedom (DoF) is considered a potential solution to address the unprecedented demand for data traffic, but there is a limit to effectively improving the communication capacity by its integer quantization. We propose a data transmission system using fractional mode encoding and deep-learning decoding. Spatial modes of Bessel-Gaussian beams separated by fractional intervals are employed to represent 8-bit symbols. Data encoded by switching phase holograms is efficiently decoded by a deep-learning classifier that only requires the intensity profile of transmitted modes. Our results show that the trained model can simultaneously recognize two independent DoF without any mode sorter and precisely detect small differences between fractional modes. Moreover, the proposed scheme successfully achieves image transmission despite its densely packed mode space. This research will present a new approach to realizing higher data rates for advanced optical communication systems.

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

“…As shown in Fig. 2 , fractional TCs give rise to local variations in the spatial field distribution, allowing a deep-learning model to discriminate them effectively despite the small mode interval 18 . Here, we assume 5 types of 10-OAM free-space optical links, where each of them has different mode spacing among .…”

“…To utilize BN effectively, we adopted the so-called pre-activation structure of BN-ReLU-Conv presented in modern network architectures such as residual network (ResNet) 40 and densely connected network (DenseNet) 41 . The ReLU is a nonlinear activation function of the form , whose purpose is to provide non-linearity to the output 18 . The ReLU is suitable for deep neural networks possessing many hidden layers because it can solve the gradient vanishing problem 39 .…”

“…Artificial neural networks comprised of multiple processing layers have been widely applied as a precise, efficient tool for detecting OAM modes 17 – 24 . Deep-learning models directly recognize transmitted spatial modes without any optical mode sorter for extracting phase information 18 , 20 and provide reliable performance against distorting factors such as optical misalignment 18 , 24 , 25 . Adaptive demodulation 21 – 23 and turbulence correction 26 , 27 for integer vortex beams have been studied in depth by pioneering researchers.…”

“…Adaptive demodulation 21 – 23 and turbulence correction 26 , 27 for integer vortex beams have been studied in depth by pioneering researchers. Meanwhile, the previous studies on fractional OAM modes focused on high-resolution recognition and the capability to increase communication capacity 17 , 18 . The researchers have experimentally demonstrated that a deep-learning model based on a convolutional neural network can classify fractional OAM modes with a resolution of up to 0.01 17 and process two independent spatial degrees of freedom simultaneously 18 .…”

“…In particular, at long-distance propagation where phase distortion accumulates, such adversarial effects become more apparent. Compared to the integer OAM mode of the ring structure, the peculiar spatial distribution of the fractional OAM mode can provide various local features for effective classification 11 – 13 , 18 . However, optical systems employing fractional OAM modes are expected to be highly sensitive to external perturbations due to tiny structural differences between adjacent modes.…”