Modulation Instability Control via Optical Seeding and Machine Learning Optimization

Spectral broadening due to Modulation Instability (MI) during nonlinear propagation in fiber optics, can arise from noise amplification and involve incoherent and complex coupled processes, making their experimental control challenging. Over the years, MI control within both noise-driven and seeded regimes demonstrated different degrees of success [1–3]. In this experimental study, we propose a new approach of MI control by all-optical seeding, whose parameters are suitably adjusted via machine learning optimization techniques. In particular, we use Genetic Algorithms (GA) to tune the wavelength and phase parameters of two weak (yet coherent) optical seeds competing with spontaneous (noise-driven) MI dynamics. This all-optical control of noise-driven nonlinear dynamics is used to optimize the correlation between two specific wavelengths in the incoherent output spectrum. Experimentally, an 80 fs laser pulse, centered at 1560 nm, is filtered using a programmable spectral filter to reshape the broadband pulse into a picosecond pump and conjointly generate two seeds with controllable parameters (i.e., wavelength λ and phase ϕ) - see Fig. 1(a). The three input signals are then amplified before entering a highly nonlinear fiber (HNLF), and the fluctuating output spectra detected in real-time using the dispersive Fourier transform (DFT) technique [4]. The corresponding spectral correlation maps can be extracted, as seen in Fig. 1 (c-d), and the correlation between two selected wavelengths $\rho(\lambda_{\mathrm{a}},\ \lambda_{\mathrm{b}})$ can be optimized by iteratively adjusting the wavelength and phase of the input seeds, as shown in Fig. 1(b).