Machine learning aided inverse design for vector modes-based weak-coupling few-mode fiber

Abstract

The vector modes (VMs) are the eigenstate modes of the optical fiber, which can be degenerated into linearly polarized (LP) modes. VMs-based weak-coupling few-mode fiber (FMF) has more advantages in communication systems and the design of these fibers is also more difficult. In this work, we present an inverse design method for VMs-based weak-coupling FMF with machine learning (ML) algorithms. The artificial neural network (ANN) and random forest (RF) are applied separately in the process of inverse designing. The lowest prediction accuracy of the two algorithms is 0.99360 and 0.99200, respectively, which shows the excellent performance of these two ML algorithms and the ANN has higher prediction accuracy than RF. Meanwhile, the inversely designed few-mode fiber can realize the degenerative separation of vector modes in the second-order and third-order mode groups, and meet the performance requirements of weak-coupling FMFs. And we obtain the effect of fiber structural parameters on mode effective refractive index (Δneff) by using feature importance ranking. In addition, the performance of the above two algorithms and the traditional optimization algorithms (Particle swarm optimization algorithm and Genetic Algorithm) in the design of weak-coupling FMFs is compared and analyzed. The proposed ML-based method provides an efficient and accurate prediction of weak-coupling FMFs, which could be greatly applied in the field of the generation of vector beam and mode division multiplexing.