Stochastic parallel gradient descent optimization based on decoupling of the software and hardware

TL;DR

This paper introduces decoupled stochastic parallel gradient descent (SPGD) models, including software and hardware approaches, which improve convergence speed and noise tolerance in correcting atmospheric turbulence phase distortions.

Contribution

The paper proposes new software and hardware decoupling methods for SPGD, enhancing convergence and noise robustness in atmospheric turbulence correction.

Findings

Hardware decoupling accelerates convergence.

Methods show strong noise tolerance.

Effective phase distortion compensation after tens of iterations.

Abstract

We classified the decoupled stochastic parallel gradient descent (SPGD) optimization model into two different types: software and hardware decoupling methods. A kind of software decoupling method is then proposed and a kind of hardware decoupling method is also proposed depending on the Shack-Hartmann (S-H) sensor. Using the normal sensor to accelerate the convergence of algorithm, the hardware decoupling method seems a capable realization of decoupled method. Based on the numerical simulation for correction of phase distortion in atmospheric turbulence, our methods are analyzed and compared with basic SPGD model and also other decoupling models, on the aspects of different spatial resolutions, mismatched control channels and noise. The results show that the phase distortion can be compensated after tens iterations with a strong capacity of noise tolerance in our model.