Research Of Stochastic Parallel Gradient Descent Based On Segmentation Random Disturbance

Compared with the conventional adaptive optics(AO), The measurement and reconstruction of wavefront information is omitted in wavefront-sensorless adaptive optics system. It serves the control signal for wavefront corrector as optimal parameters. The approximate ideally correction effect is obtain by optimizing and implementing the performance index. The characteristics, such as the simple hardware implementation, without the measurement and reconstruction of wavefront information, and the strong ability to adapt in the complex conditions, expand the application space of the conventional AO technology.Wavefront-sensorless adaptive optics system based on stochastic parallel gradient descent algorithm and the Stochastic Parallel Gradient Descent Based on Segmentation Random Disturbance are the main research content of this article. The original rational research of the SPGD algorithm are processed by using the numerical simulation. Then, the correction characteristic of static atmospheric turbulence of wavefront by adopting the traditional SPGD algorithm is verified furtherly.Meanwhile, the main factors of the traditional SPGD algorithm convergence properties are analysed in Details. Then, different values are taken in the optimal value range of the initial random perturbation amplitude respectively, and The corrections for static atmospheric turbulence distortion by adopting the traditional SPGD algorithm are implemented under gain coefficient 0.85. The simulation analysis show that traditional SPGD algorithm obtains the optimal convergence characteristics under the random perturbation amplitude 0.2.Firstly, the convergence characteristics is relatively slow based on the detailed analysis of SPGD algorithm. Thus, the Stochastic Parallel Gradient Descent Based on Segmentation Random Disturbance is put forward in the artilcle. Secondly,wavefront-sensorless adaptive optics system is built with a 61-element deformation mirror and implements the correction of wavefront aberrations, which is simulated by the 65-order Zernike polynomials and meets the Kolmogorov spectrum. Compared with the best fixed initial perturbation amplitude SPGD algorithm, the SR convergence characteristics increases 1.6 times by adopting the SPGD algorithm based on the segmentation random perturbation amplitude and the best segmentation number isL =4(5*c).Meanwhile, the optimal value range of the optimal initial random disturbance amplitude is the researched and analysed under the interval 0.01,the premise of section 80. Then, it can be concluded that the best value range of initial random perturbation amplitude is 0.3 ~ 1.5 rad. Finally, the optical experimental platform is constructed, and the correction ability of AO system is verified furtherly.