In high-speed optical communications, traditional orbital angular momentum (OAM) multiplexing systems face fundamental limitations, including exponentially increasing spatial-domain complexity, aggravated modal crosstalk, and strong dependence on continuous-wave lasers. These challenges hinder scalability and robustness in complex environments.
To address this, a research team led by Professor Fu Feng and Professor Xiaocong Yuan from Zhejiang Lab has developed a novel OAM-based spatiotemporal multiplexing (OAM-STM) architecture. This approach couples pulsed OAM beams with a diffractive deep neural network (D 2 NN) and optical fiber delay-line arrays, establishing a “space encoding–time decoding” transmission link. In this design, pulsed OAM states are spatially separated into distinct “activation regions” by the D 2 NN and then mapped into the time domain via fiber delay lines before being detected by a single-pixel photodetector.
The experimental demonstration achieved 3-bit data transmission using ultrafast pulsed lasers with 10 ps pulse width. A digital micromirror device (DMD) generated OAM beams carrying binary data patterns (“001” to “111”) with topological charge l ∈ [1, 3]. After D 2 NN modulation, the beams were focused into three activation regions and transmitted through fiber delays (2 m, 4 m, 6 m), producing distinct temporal pulse sequences with a 9.48 ns delay difference. By applying an intensity threshold of 0.6, the system accurately decoded the 3-bit data.
Although the experimental system operated at kHz rates due to DMD switching speed (10,752 Hz), the OAM-STM architecture is inherently compatible with high-repetition-rate OAM sources. In principle, its demultiplexing speed is limited only by photodiode bandwidth, enabling scalability to the GHz regime. Increasing D 2 NN layer count or neuron density can further enhance bit capacity.
This breakthrough overcomes the low time-domain utilization and high demultiplexing complexity of conventional OAM communications. By integrating a temporal multiplexing dimension into an all-optical framework, each laser pulse can carry multiple times more data, alleviating the throughput bottleneck caused by pulse repetition rate limits. Furthermore, all-optical decoding avoids the latency and losses of electronic signal processing, enabling more compact and efficient high-speed systems.
Looking forward, the researchers envision upgrades in three directions: (1) adopting high-repetition-rate pulsed OAM generators and lasers to directly achieve GHz-level transmission rates; (2) developing multilayer D 2 NN and on-chip integrated delay lines (e.g., high-index spiral waveguides) for miniaturized devices suitable for 5G all-optical networks; and (3) extending the technique to more challenging environments such as long-distance free-space links, underwater optical wireless channels, and quantum communications.
This work provides a new paradigm for spatiotemporal multiplexing in optical communications, fostering deeper integration between all-optical neural networks and OAM technologies, and paving the way toward next-generation high-capacity, high-adaptability optical information systems.
High-speed all-optical neural networks empowered spatiotemporal mode multiplexing