Seq2Seq model with attention for predicting nonlinear propagation of ultrafast pulses in optical fibers

Abstract

The propagation of ultrashort pulses in optical fibers exhibits highly complex nonlinear dynamics, which plays a central role in the development of light sources and photonic technologies. The mainstream of the existing methods for modeling and predicting complex nonlinear propagation dynamics of ultrashort pulses in optical fibers is based on recurrent neural networks (RNNs), which use a Multi-In-Single-Out (MISO) architecture to predict the optical pulse evolution recursively. This autoregressive model is severely limited by the error accumulation problem and also requires significant computational resources. Affected by the error accumulation problem, this method often leads to severe performance degradation in long sequence prediction tasks, thus limiting the practical application of the prediction model. In this work, we propose a new non-autoregressive model using a Single-In-Multi-Out (SIMO) architecture to simulate the highly nonlinear dynamics of ultrashort pulse propagation in optical fibers. Our model is validated on the public dataset. The results show that our model can significantly reduce the prediction error in modeling and predicting the complex nonlinear propagation of ultrashort pulses in optical fibers. In addition, the required computational resources and time spent are significantly reduced. As a whole, our proposed method comprehensively outperforms the mainstream methods in terms of efficiency, accuracy and practicality. We believe our work could bring new insights into the modeling and analysis of complex ultrafast nonlinear dynamics.