Multi-Mode-Group-Equalized Fiber Designed via a NN-GA Framework Based on Mode Distribution Figure of Merit

In short-distance high-capacity optical communication scenarios, space-division multiplexing technology based on multimode fiber enables large-capacity transmission at lower costs by efficiently utilizing mode resource, showing significant application potential. However, conventional multimode fibers are plagued by severe mode group (MG) crosstalk and significant differential mode delay (DMD), which severely undermine signal integrity and increase bit-error rates, thus limiting their practical use. This paper proposes a multi-mode-group-equalized fiber (MGEF) characterized by the equalization of crosstalk and DMD. Introducing the mode distribution figure of merit (MD-FOM) as a criterion, the MGEF is designed by integrating neural network and genetic algorithm (NN-GA) and adopting a complex multi-ring core structure for fiber design. MD-FOM guides NN-GA optimization, with larger values indicating better alignment with design requirements, enabling iterative refinement of fiber structure for desired mode separation and equalization performance. The MGEF can support over 100 modes, which are divided into 10 MGs according to effective refractive index (neff), and the minimum difference of neff(Δneff) between MGs is 1.52×10-3. It exhibits excellent performance with inter-group crosstalk lower than -20 dB/km, maximum DMD under 15 ps/m, and the fabrication tolerance for the radius and refractive index difference of each layer in the multi-ring core is ±3%. This effectively addresses the aforementioned critical issues.