High-fidelity optical wireless transmission in complex environments around a corner using the design of a single-layer neural network for data encoding.

In this paper, an optical data encoding scheme is proposed to realize high-fidelity optical wireless transmission (OWT) through complex media (i.e., dynamic and turbid water) around a corner using a single-layer neural network to fit a physical model. The physical process of optical modulation, wave diffraction, and single-pixel optical detection is modeled as a convolution operation in the designed single-layer neural network. Each pixel of a 2D image to be optically transmitted can be encoded into a random pattern without the usage of any datasets or labels in the neural network. The designed single-layer neural network offers a simplified structure and enables fast and straightforward 2D pattern generation. The generated random patterns serve as information carriers to control optical waves in a free-space optical channel. At the receiving end, a series of light intensities are collected by a single-pixel detector (SPD). Optical experiments are extensively conducted, and it is demonstrated that the developed system can realize high-fidelity and high-robustness optical data transmission through dynamic and turbid water around a corner under various conditions, e.g., different water turbidities and different separation distances around the corner. It could be believed that the proposed method can provide insight into neural networks via the fitting of a physical model for the OWT and facilitate a wide range of real-world applications in complex environments.