Geração de pentes de frequência ópticos programáveis via design inverso e deep learning.

Abstract

Optical frequency combs (OFCs) are discrete sets of continuous wave carriers periodically spaced in frequency that have important applications in measurement, communication, spectroscopy and other fields. Generally, OFCs need to be designed according to different applications. However, existing methods for designing the operating parameters of OFC generators, such as those based on optimization algorithms, are time-consuming, inefficient and difficult to achieve optimal results. In this work, the Inverse Design method of OFC using Deep Learning neural networks is experimented with, which produces results in a timely manner compared to conventional methods and can also improve the performance of the generated OFC. In this method, according to the target OFC required, the trained neural network can be used to inversely design the corresponding parameters. The method is applied to two cases of OFC generators based on cascades of electro-optical modulators (EOC) with a single control signal. The trained neural networks are tested under the same conditions and compared with each other. The implemented method can be extended to more complex OFC generators, making it a promising tool for the efficient design of OFCs.