On distributed wavefront reconstruction for large-scale adaptive optics systems

Abstract

The D-SABRE (Distributed Spline based ABeration REconstruction) method is proposed for distributed wavefront reconstruction with application to large scale adaptive optics systems. D-SABRE decomposes the wavefront sensor domain into any number of partitions and solves a local wavefront reconstruction problem on each partition using multivariate splines. D-SABRE reconstruction accuracy is within 1% of a global approach with a speedup that scales quadratically with the number of partitions. The D-SABRE is compared to the CuRe-D method in open-loop and closed-loop simulations using the YAO adaptive optics simulation tool. D-SABRE reconstruction accuracy exceeds Cure-D for low levels of decomposition, and D-SABRE proved to be more robust to variations in the loop gain.

Distributed approach for Aerodynamic Model Identification of the ICE aircraft using multivariate splines

“…By using the barycentric formulation of the polynomials, the basis is stable and shows easy formulations to enforce any order of continuity between the subdomains [10]. This model structure has been successfully applied to AMI, but also to other fields, such as adaptive optics or partial differential equations [8,11,12].…”

“…However, all these methods still solve a global problem, not fully exploiting the local properties of the simplotope splines. Distributed methods have been used as well with B-Splines, using Dual Ascent or the Alternating Direction Method of Multipliers (ADMM) [12,20]. However, this has only been used for the 2-dimensional wave front reconstruction problem and has not been applied to the higher dimensional AMI case yet.…”

GPU implementation for spline-based wavefront reconstruction

“…However, D-SABRE is subject to propagation of errors that are created in the estimation of the piston offsets between partitions, if a high level of domain composition is applied or if the domain contains large internal obscurations. Applying overlap between the partitions mitigated but did not negate this effect and also decreases the computational speed [10]. This phenomenon yields a trade-off in WF accuracy and number of partitions G, putting a limit on the latter, which prevents the D-SABRE method from reaching its full potential of linear or even sublinear computational load per processor for very or extremely large-scale AO systems.…”

“…The distributed-spline-based aberration reconstruction (D-SABRE) method [10] was recently introduced as an extension of the spline-based aberration reconstruction (SABRE) method [11]. The approach uses multivariate simplex B splines [12] to locally model WF aberrations and allows application on non-rectangular WFS arrays.…”

“…The D-PME is an iterative process that requires only communication between neighboring partitions. The overall method has a theoretical computational complexity of ON 2 ∕G 2 flops (floating point operations), which have to be performed per parallel processor, for a total of G partitions and scales, therefore linearly with the number of WFS measurements for G ≥ ffiffiffiffi ffi N p [10]. The D-SABRE method was extensively compared to the Cumulative Reconstructor with domain decomposition (CuRe-D) method [14], a line integral approach with domain decomposition that has linear computational complexity and is suitable for parallel implementation.…”