Enhancement of Reservoir Computing Performance via Ultrashort Optical Feedback

Abstract An enhanced delay-based photonic reservoir computing system assisted by ultrashort optical feedback is proposed and numerically demonstrated. Specifically, a picosecond-scale ultrashort optical feedback loop is integrated into a conventional semiconductor laser-based single-feedback reservoir computing (RC) architecture. The ultrashort feedback loop acts as a fast dynamical regulator, enabling multiscale temporal dynamics while largely preserving the structural simplicity of the single-feedback configuration. The RC performance is evaluated in terms of separability and consistency. The results show that appropriately selected ultrashort feedback parameters can increase between-class separation and reduce within-class dispersion while maintaining reproducible reservoir responses. Moreover, the memory capacity performance and benchmark evaluations using Santa Fe chaotic time-series prediction and nonlinear channel equalization demonstrate that the proposed system outperforms the conventional single-feedback reservoir.