image: (a) Neural network architecture; (b) nonlinear Schrödinger equation simulation for the data generation and analysis; (c) experimental setup of the mode-locked fiber lasers; (d) comparison of prediction and experimental results from the autocorrelation phenomena of two-color solitons.
Credit: ©Science China Press
Multicolor fiber lasers—lasers that produce light at multiple colors (wavelengths) at once—are gaining attention for their wide-ranging uses in areas like high-speed communications, laser-based measuring tools, and ultra-sensitive sensors. But beyond their practical value, they also offer a fascinating window into the world of solitons—special light pulses that can travel long distances without changing shape.
Inside these lasers, solitons don't just move along quietly—they can collide, merge, switch states, or even explode in chaotic bursts. These behaviors are part of a complex and highly nonlinear world that has traditionally been very difficult and slow to simulate using standard math-based models.
To tackle this, researchers have turned to artificial intelligence. In a recent breakthrough, researchers from Zhejiang A&F University developed a smart neural network model that combines two types of deep learning algorithm: convolutional neural networks (CNNs), which are good at spotting patterns in space; and recurrent neural networks (RNNs), which excel at understanding how things change over time. The researchers combined the two methods and designed a dual-channel model, which can process the real and imaginary components of the optical field separately but synchronously. This approach not only addresses the limitations of standard deep learning frameworks in handling complex-valued data, but also maintains the physical coherence of optical field interactions, enabling highly accurate predictions of multicolor soliton dynamics.
As a result, this dual-channel system can accurately predict how two-color and three-color solitons will evolve—even in tricky situations like unstable states or during intense collisions. It can track changes in their energy, wavelength, and phase with remarkable accuracy, all while matching up well with both computer simulations and real-world experiments.
This work not only pushes the boundaries of deep learning in physics but also offers a powerful new tool for understanding and controlling the wild dynamics inside multicolor lasers—potentially leading to better laser technologies in the future.
Journal
Science China Physics Mechanics and Astronomy
Method of Research
Computational simulation/modeling