Adaptive wavefront control with asynchronous stochastic parallel gradient descent clusters

Abstract

A scalable adaptive optics (AO) control system architecture composed of asynchronous control clusters based on the stochastic parallel gradient descent (SPGD) optimization technique is discussed. It is shown that subdivision of the control channels into asynchronous SPGD clusters improves the AO system performance by better utilizing individual and/or group characteristics of adaptive system components. Results of numerical simulations are presented for two different adaptive receiver systems based on asynchronous SPGD clustersone with a single deformable mirror with Zernike response functions and a second with tip-tilt and segmented wavefront correctors. We also discuss adaptive wavefront control based on asynchronous parallel optimization of several local performance metrics-a control architecture referred to as distributed adaptive optics (DAO). Analysis of the DAO system architecture demonstrated the potential for significant increase of the adaptation process convergence rate that occurs due to partial decoupling of the system control clusters optimizing individual performance metrics.

Wavefront spatial‐phase modulation in visible optical communications

“…As final remark, taking into account that the turbulence imposes a constraint to this modulation scheme, the use of adaptive algorithms can mitigate the turbulence‐induced wavefront distortion effect …”

Geometrical optics analysis of atmospheric turbulence

“…Though in experiments the refractive index is not a direct measurable parameter, one can always use temperature and pressure sensors to extract relative refractive index deviations, which serves as a practical way for finding inputs for a simulative estimation of the channel. For a Gaussian beam, similar methods have been applied and Fig 8-11 are turbulence effects on coherent beam, partial coherent beam (like diode laser signal), random walk fitting and equivalent C n 2 according to equation (14). Though the simulation ignores the small eddies refraction behaviors, it gives a good estimation of the turbulence level and its distortion on the beam.…”

“…Based on the theoretical results of equation (14) and equation (15), both would converge to the same dominant z 3 term in linear portion to C n 2 if the propagation distance is long enough. With the help of deformable mirrors, one can intentionally pre-distort the beam a little by adding regional tilts through guidance from initial beam distortion, and the resultant "random walk" would be reduced and lead to an equivalently smaller C n 2 .…”

Bandwidth analysis and improvement of the beam phasing of fiber amplifiers using the stochastic parallel gradient descent algorithm

“…D-SPGD algorithm incorporates features of both wave-front sensor-based adaptive optics and model-free gradient descent optimization wave-front control, which offers fast adaptation convergence even for high-resolution adaptive optics. In [24] it is shown that subdivision of the control channels into asynchronous SPGD clusters improves the AO system performance by better utilizing individual and/or group characteristics of adaptive system components. Following the idea of defining vector cost function proposed in [18] and [25], in the present section, we propose cross-grouped SPGD algorithm, in which cost function information is transferred unidirectional, and also a few more detectors are required.…”