High-speed all-optical neural networks empowered spatiotemporal mode multiplexing.

Abstract

Orbital angular momentum (OAM) beams, characterized by a helical phase structure and phase singularity, have emerged as a powerful resource for high-capacity optical communications through mode-division multiplexing (MDM). Traditional OAM multiplexing systems operating solely in the spatial domain face significant challenges, including increased system complexity, inter-modal crosstalk, and limited scalability. Recent advances have explored hybrid multiplexing schemes combining OAM with wavelength or polarization degrees of freedom, demonstrating Pbit/s level transmission capacities. However, these systems predominantly rely on continuous-wave lasers and external modulators, which constrain their applicability in challenging environments, whereas pulsed lasers provide superior peak power, enhanced transmission robustness, and the potential for implementation of OAM lasers, which generally emit pulsed OAM beams. Here, we report an OAM-based spatiotemporal multiplexing (OAM-STM) technique that synergistically implements pulsed OAM beams with a diffractive deep neural network (D2NN) and optical fiber delay lines to project spatial mode information into the temporal domain. This approach leverages the full potential of pulsed laser sources by activating the underutilized time dimension, thereby overcoming the repetition-rate bottleneck and enhancing channel throughput. We experimentally demonstrate an OAM-based spatiotemporal demultiplexer achieving demultiplexing speed limited only by the bandwidth of the photodiode if OAM generation is fast enough. In the meantime, the architecture is intrinsically compatible with high-repetition-rate OAM sources, offering the entire system the scalability to GHz rates. This work establishes a foundational framework for high-speed, all-optical, and high-capacity OAM-STM systems, with promising implications for free-space optical communication, underwater communication links, and other complex environments.