基于低复杂度Transformer的光纤信道快速精确建模技术

摘要：光纤信道建模对于表征光纤特性、开发先进数字信号处理算法十分重要。基于物理模型的分步傅里叶算法（SSFM）需要大量迭代运算，复杂度较高，限制了其应用前景。提出了一种基于低复杂度Transformer架构的光纤信道建模方法。在传统Transformer架构的基础上，我优化绝对位置编码为相对位置编码，优化全局注意力机制为滑动窗口注意力机制，进一步增强模型对光纤非线性特征的建模效果。结果表明，所提出的方法与SSFM之间的有效信噪比（ESNR）误差仅有0.15 dB，计算时间相比传统Transformer降低69.9%，相比SSFM降低96.9%，验证了其具备较高的精度且计算复杂度大幅降低。

关键词：光传输系统；光纤信道建模；Transformer

Abstract: Optical channel modeling is essential for characterizing fiber transmission characteristics and developing advanced digital signal processing (DSP) algorithms. The split-step Fourier method (SSFM), as a physics-based numerical solver, provides high accuracy but suffers from high computational complexity due to extensive iterative computations, which limits its practical deployment. A low-complexity Transformer-based architecture is proposed for optical fiber channel waveform modeling. Based on the standard Transformer framework, two key modifications are introduced: (1) replacing absolute positional encoding with relative positional encoding, and (2) substituting global attention with a sliding-window mechanism, thereby enhancing the model’s ability to capture complex nonlinear behavior. Results demonstrate that the proposed method achieves an effective signal-to-noise ratio (ESNR) error of only 0.15 dB compared to SSFM, while reducing inference time by 69.9% relative to the standard Transformer and by 96.9% compared to SSFM. These results validate the proposed approach as both highly accurate and computationally efficient, offering a promising solution for fast and accurate optical channel modeling.

Keywords: optical transmission system; optical fiber channel modeling; Transformer