Deep Residual Learning Recognition of Orbital Angular Momentum Superpositions Over Oceanic Turbulence Using a Cubic Phase

Orbital angular momentum (OAM) is an important property of light that has attracted considerable attention for increasing the system capacity of optical wireless communications. In undersea communication scenarios, the presence of ocean turbulence distorts the propagating beams and thereby degrades the OAM recognition performance. In order to realize high-fidelity OAM-based data transmission, we propose a deep-residual-learning-based scheme to identify the oceanic-turbulence-distorted OAM modes. The proposed scheme not only shows remarkable recognition accuracy on both single and superposed OAM basis, but also exhibits strong extrapolation ability to predict unseen samples between or beyond the training samples. Moreover, our scheme can work on more kinds of superposed OAM modes owing to a cubic phase, added right before the neural network, that solves the degeneracy issue of OAM superposition. Finally, simulations of transmitting a cat image encoded by OAM modes yields high fidelity, which demonstrates the utility of our design.