Experimental Validation and Performance Analysis of Photonic Reservoir Computing Using Multi-Delay Coupled DFB Lasers for Optical Chaotic Signal Separation

This paper addresses a common framing in prior photonic reservoir computing (PRC) work: the relationship between a reservoir laser's self- feedback delay ($\tau _{s}$) and target sampling period ($T$) is often characterized as a choice—specifically, configurations like$\tau _{s}=T$or$\tau _{s}=T+\theta$(where$\theta$denotes the virtual node time interval). It can lead to suboptimal performance for open-loop tasks (e.g., signal separation) due to delay clock-time resonance effects. In this paper, we propose a novel PRC architecture using multi-delay coupled short-cavity distributed feedback semiconductor lasers (SC-DFB-SLs), which leverages a ring-coupled network to form three parallel reservoirs via the self- feedback and mutual coupling cavities of two SC-DFB-SLs. This architecture is explicitly tailored for the task of separating three linearly mixed optical chaotic signals, providing clear context for interpreting performance metrics. The theoretical model of the proposed system is built on the Lang-Kobayashi equations, and validated experimentally with a custom-built multi-delay coupled PRC setup. Key findings reveal that the architecture eliminates the need for the commonly used$\tau _{s}=T$configuration: optimal signal separation performance is jointly determined by$T$and the coupling coefficient ($k_{c}$) of the SC-DFB-SLs, with$\tau _{s}$framed as an independently tunable parameter. For the targeted optical chaotic signal separation task, the theoretical normalized mean square error ($NMSE$) reaches as low as 0.052, while the experimental$NMSE$is 0.084. The system also achieves a memory capacity ($MC$) of up to 23.4, which significantly outperforms single-delay PRC architectures, such as those based on a single DFB-SL or vertical-cavity surface-emitting laser (VCSEL), where reported memory capacities typically fall below 15. This research clarifies the current state of knowledge on delay clock-time relationships in PRC, provides a tunable framework for designing high-dimensional nonlinear PRC systems, and holds potential for applications in optical communication and chaos-based security.