Groundbreaking experimental demonstration of generalized chaos synchronization prediction in multi-drive unidirectionally-coupled laser systems using DFB-based photonic reservoir computing

Abstract

Photonic Reservoir Computing (PRC) has emerged as a powerful tool for complex photonic dynamics, offering distinct advantages over traditional methods that often struggle with causality modeling in unidirectionally coupled systems. In this study, we experimentally explore a PRC physical hardware system based on a delay-feedback distributed feedback (DFB) laser, uncovering its remarkable ability to learn unidirectional coupling schemes and predict the response dynamics of multi-drive systems using limited time-series data from the drive-response DFB laser system. After training with partial dynamics data from the drive-response semiconductor laser system, the PRC demonstrates precise prediction capability for the dynamics of the response DFB laser system across various drive optical signals sharing the same coupling scheme. Notably, even when the drive laser system is replaced, the PRC effectively reproduces the dynamics of the response laser system by leveraging the dynamics of the new drive system. Furthermore, the trained reservoir achieves high-quality synchronization with outputs from different drive-response systems. Our analysis delves into the effects of sampling period (T = 30 - 60 ns) and the number of virtual nodes (N = 50 - 300) on the normalized mean square error (NMSE), while also confirming the robustness of feedback strength ($K_f$ = 0.1 - 0.2) and injection strength ($K_{inj}$ = 0.06 - 0.2) to synchronization quality. Experimental results reveal that the PRC system consistently achieves high-quality chaotic synchronization (correlation coefficient ρ > 0.93) across various drive modes, with a prediction error NMSE less than 0.133. Particularly, when T = 60 ns and N = 300, the NMSE of the drive-response system, constructed using optical feedback DFB lasers and optical injection DFB lasers, drops to as low as 0.0951, underscoring the efficacy of parameter optimization. This research highlights the generalization capability of PRC under complex drive signals, paving the way for a new paradigm in dynamics prediction for multi-physical field coupled systems. The findings hold significant promise for advancing applications in photonic computing and beyond.