All-optical implementation of an optoelectronic oscillator reservoir computer with an integrated spatial photonic...

Reservoir computing (RC), inspired by neural networks, has gained significant attention for tackling complex tasks such as recognition and classification. Most RC implementations are limited to software, which restricts processing speed, as reservoir outputs are often processed through slow, digital offline post-processing in previous work. To address these constraints, we propose a solution that combines a nonlinear time-delayed optoelectronic oscillator (OEO) with an integrated spatial photonic processor (ISPP) for online reservoir output processing. This approach leverages both temporal multiplexing and spatial processing, enabling high-speed, and parallel hardware implementation. We achieve enhanced performance in modulation format recognition (MFR) for visible light communication (VLC) signals using a photonic reservoir with 64 nodes. Tests on 16 widely used modulation formats at a 2 Gbps data rate over a 1.2 m VLC transmission link demonstrate that this spatiotemporal multiplexing technique can accurately classify all formats, achieving a state-of-the-art overall accuracy of 99.5%.