Deep Learning Waveform Channel Modeling for Wideband Optical Fiber Transmission: Model Comparisons, Challenges and Potential Solutions

Fast and accurate waveform simulation is critical for characterizing optical fiber channel behavior, developing digital signal processing (DSP) algorithms, optimizing optical network configurations, and advancing optical transmission systems towards wideband. Deep learning (DL) has emerged as a promising approach for waveform modeling, offering accuracy comparable to the traditional split-step Fourier method (SSFM) with significantly lower computational complexity. Despite substantial progress, most existing DL schemes are designed for few-channel, low-rate wavelength-division multiplexing (WDM) systems. Moreover, a lack of standardized accuracy evaluation methods hinders fair comparison among DL schemes, slowing their development for wideband applications. In this paper, we propose a DSP-assisted accuracy evaluation method integrated with nonlinear DSP, which provides a fair benchmark for assessing modeling accuracy. Using this method, we conduct a comprehensive comparison of DL schemes, spanning from simple few-channel, low-rate configurations to complex multi-channel and high-rate scenarios. The results demonstrate that feature decoupled distributed (FDD)-based schemes combining with temporal neural networks achieve superior modeling accuracy, especially under these more complex conditions. Despite their advantages, we observe that even FDD-based models experience accuracy degradation as the number of channels and symbol rate increase. We analyze the underlying challenges related to sampling rate, dispersion-induced interference, and nonlinear effects. Potential solutions are discussed, including the integration of additional physical priors and structural optimization of DL models. We believe that the standardized evaluation framework, comparative results, and in-depth challenge analysis presented in this work will facilitate the rapid advancement of DL-based channel modeling in wideband systems and support the evolution of next-generation optical networks.