Hybrid PINN-TWM framework for DFB laser modeling and parameter estimation under small-sample constraints

Accurate modeling and parameter estimation of distributed feedback (DFB) lasers are limited by the strong nonlinearity of laser dynamics and small-sample constraints. Based on the traveling-wave model (TWM), a hybrid physics-informed neural network (PINN)-TWM framework is proposed to improve data efficiency and enable accurate DFB laser response modeling. The framework maps intrinsic laser parameters to light-current (P-I) curves and small-signal frequency responses. For parameter estimation, a cascaded forward-inverse PINN architecture is developed, in which the trained physics-informed forward surrogate provides response-level guidance for recovering intrinsic physical parameters. Simulation results show that the forward model consistently outperforms the convolutional neural network (CNN) baseline across the evaluated device responses, reducing the MSE by more than 40% under the same limited training set of 500 samples. Meanwhile, the cascaded framework achieves the lowest parameter estimation error among the compared inverse methods. This study demonstrates the potential of combining physics-based modeling with data-driven learning for DFB laser response prediction and parameter estimation under small-sample conditions.