Artificial intelligence–driven optimal control and bifurcation avoidance in semiconductor laser systems

Engineering notes

Key topics: autonomous driving, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Injection-locked semiconductor lasers feature a complex nonlinear dynamic response due to the coupling between the carrier density and optical field in the presence of an external driving force. Although this class of systems has been used to achieve many interesting photonic capabilities, they are quite unstable and prone to Hopf bifurcations, leading to self-oscillations that undermine their effectiveness in applications where stable operation is critical. In this study, a full computational framework is constructed to investigate and control the dynamics of injection-locked semiconductor lasers. A dimensionless system is established using coupled rate equations to describe the physical processes within it. The stability boundaries and Hopf bifurcations are determined by performing bifurcation analysis using the MATCONT software package. To incorporate stability constraints into the optimization algorithm, a neural network surrogate is developed to predict the maximum real part of the system’s eigenvalues as a function of both state variables and control parameters. A soft penalty-based method is used to avoid operation close to an unstable region, thus imposing stability without any non-smooth restrictions on the system. The corresponding optimization problem is solved by employing IPOPT. It is shown that the developed method suppresses oscillatory behavior, producing smoother control inputs with better objective function performance than no control at all. This combination of techniques yields a versatile numerical scheme for stabilization of nonlinear photonic systems, which can be applied to a large variety of systems with instability caused by bifurcations.

Links and sources

• Official / arXiv page

Comments