Experimental demonstration of all-optical input masking in Photonic time-stretch reservoir computing

We propose and experimentally demonstrate a photonic reservoir computing (RC) system with integrated all optical input masks to enhance processing performance. Leveraging photonic time-stretch (PTS) and spectral mixing, the system employs optical wavelengths as virtual reservoir nodes, thereby improving spectral interaction and memory capacity during feedback. To overcome electronic bottlenecks and reduce bandwidth requirements, we design two optical input mask schemes: (i) a programmable optical filter based on parallel diffraction gratings, and (ii) an analog mask generated by a Mach Zehnder Interferometer (MZI). Both approaches introduce randomness directly in the optical domain prior to modulation, enriching reservoir dynamics without relying on high-speed electronics. Proof-of-concept experiments on waveform and spoken-digit classification validate the effectiveness of our method. The proposed system achieves accuracies of 97.19% with the MZI based mask and 98.16% with the diffraction gratings mask, compared with 80.65% in the no-mask case. These results demonstrate that all-optical masking not only significantly improves classification performance but also alleviates electronic constraints, making it highly suitable for real-time recognition in high-speed photonic imaging systems.