UVL-based hybrid neural network model for generalized optical fiber channel modeling

Accurate fiber channel modeling is essential for the evaluation of optical communication systems, but conventional methods like the split-step Fourier method (SSFM) are hugely time-consuming. Although neural networks (NNs) such as Transformer or bidirectional long short-term memory (Bi-LSTM) can accelerate the modeling, it is still complex and time consuming for multi-channel transmissions. Besides, system parameters such as the number of transmitted channels, distance or launch power may be varied, so the NNs should have strong generalization ability. In this paper, we propose a universal virtual lab (UVL)-based hybrid neural network modeling scheme that takes the advantages of fast multi-channel nonlinear interference noise (NLIN) calculation and fitting ability to achieve accurate and generalized modeling for multi-channel transmissions. Simulation results show that the proposed model maintains signal-to-noise ratio (SNR) errors within 0.5 dB of SSFM benchmarks while enabling over 550× speedup for 800-km wavelength-division multiplexing (WDM) transmission with 17 channels. This hybrid approach significantly improves the modeling flexibility and efficiency for multi-channel optical fiber communication systems.