Principle driven parameterized fiber model based on GPT-PINN neural network

Abstract

To cater the need of Beyond 5G communications, large numbers of data driven artificial intelligence based fiber models have been put forward as to utilize artificial intelligence’s regression ability to predict pulse evolution in fiber transmission at a much faster speed compared with the traditional split step Fourier method. In order to increase the physical interpretabiliy, principle driven fiber models have been proposed which inserts the Nonlinear Schödinger Equation into their loss functions. However, regardless of either principle driven or data driven models, the whole models need to be re-trained under different transmission conditions. Therefore, those models’ complexity could be larger and computation costs could be higher, especially when dealing with large numbers of different transmission conditions. In order to address this problem, we propose the principle driven parameterized fiber model in this manuscript. This model breaks down the predicted NLSE solutions with respect to different transmission conditions into the linear combinations of several eigen solutions which were outputted by pre-trained eigen solution solvers. Therefore, the model can greatly alleviate the heavy burden of re-training since only the linear combination coefficients need to be found when changing the transmission condition. Not only strong physical interpretability can the model possesses, but also higher computing efficiency can be obtained. The model’s performance is demonstrated by predicting the propagating evolution of pulses with different shapes under 1000 different transmission conditions over the maximum 100 km fiber. Under both theoretic analysis and numerical demonstration, after appropriately trained, this model is able to predict pulses evolution under 1000 different transmission conditions with only 10 selected eigen solutions. Besides, the model’s computational complexity is only 0.0113 % of split step Fourier method and 1 % of the previously proposed principle driven fiber models.