Fast and high performance control


Developing smart algorithms for wavefront control

Click here for Extended Information

esfri_brunner_project4_Fast_and_high_performance_control
Recapitulation

Goal

The goal of this project is to design, implement and validate a fast and high-performance control algorithm on multi-core off-the-shelf hardware to control a large-scale adaptive optics system for application in extremely large telescopes.

 

Approach

This project will be organized but not strictly implemented in the following phases:

Phase 1: Familiarization with the subject for the PhD Student.

Phase 2: Study of the Wavefront Sensing process and development of distributed Wavefront Reconstruction.

Phase 3: Development and optimization of computational complexity of the control algorithms versus performance trade-off.

Phase 4: Implementation and validation of the control algorithm possibly on a demonstrator.

Phase 5: Real-time validation on a test bench.

Phase 6: write and defend thesis


Involved Partners
The work is mainly done by Delft Center for Systems and Control, in collaboration with staff members of other faculties and a potential role for TNO.


Progress

Goal

Adaptive Optics (AO) are used in large ground-based telescopes to compensate for disturbances in the atmosphere as well as vibrations of the telescope's optical components (see Figure 1). With the increasing dimensions of telescopes, up to 39m for the European Extremely Large Telescope (E-ELT), the dimensions of AO systems are increasing such that the computational complexity of the controller prohibits an (unstructured) centralized implementation.

 

The goal of this PhD project is to design, implement and validate a fast and high-performance control algorithm on multi-core hardware to control a large-scale AO system for application in extremely large telescopes. With smart algorithms, distributed solutions can be found which reduce the computational complexity to control the E-ELT AO systems and allow real time correction using off-the-shelf hardware like GPUs (see Figure 2).



ao

Figure 1 AO control system with:

WFS (Wave Front Sensor) = device  to measure the distortion of the light (wavefront)

DM  (Deformable mirror) =  adjustable mirror that corrects the wavefront to obtain a non-moving and sharp image

Beam splitter = a semi-permeable mirror which lets a portion of the light  through to the detector and  reflects the other  part to the wavefront sensor


gpu

Figure 2: Computational demand for E-ELT AO system can be met through massive parallelization and implementation on GPUs.


In order to achieve this goal, both the spatial and temporal dynamics have to be taken into account. The work on Spline base Aberration Reconstruction (SABRE) for SH slope measurements, which was developed at by Delft Center for Systems and Control (DCSC), will be continued. One part of the study is to fully exploit the information given by the CCD intensity images through accurate modeling of the nonlinear imaging process instead of using approximate measurements of the local gradients, in order to obtain increased accuracy for the spatial reconstruction of the wavefront (WFR). Further, the project will investigate how the wavefront can be approximated by nonlinear basis functions with a tolerable computational load. The SABRE method has been shown to allow highly parallelizable computation of the wavefront estimate by local reconstruction on a decomposition of the sensor domain and distributed optimization requiring communication only with the neighboring sub-domains. (see Figure 3) This distributed WFR method can then be combined with distributed control methods to close the feedback loop and obtain an adaption for real-time application. In this second major study, the distributed SABRE is planned to be validated in realistic simulators and with real-life data.

 

distribute

Figure 3: Partitioning of the wavefront reconstruction domain in SABRE into several sub domain run on separate GPU cores.



The PhD student will optimize the computational complexity of the resulting control algorithm versus performance trade off, and implement and validate the method on a multi-core AO demonstrator system. Further the study envisages collaborations with international teams working on AO for the new generation of extremely large-scale telescopes, as well as laboratory and on-site validation of the results.



Approach

This PhD project started the 01/05/2013 and is planned to be finished by 31/04/2017. The project will be organized but not strictly implemented in the following phases:

Phase 1: (6 months) Literature study on Control Theory, Optimization and Estimation methods, Parallel Computing, Fourier Optics and AO systems for astronomical telescopes. Familiarization and continuation of the SABRE project.

Phase 2: (9 months) Study of the Wavefront Sensing process and development of distributed Wavefront Reconstruction algorithms based on slope and intensity measurements.

Phase 3: (9 months) Distributed control methods for determining the settings of the deformable mirror based on the wavefront reconstruction. Optimization of computational complexity of the control algorithms versus performance trade-off.

Phase 4: (9 months) Implementation and validation of the control algorithm on multi-core off-the-shelf hardware and software architectures as CUDA from NVIDIA . Set up of an multi-core AO demonstrator for large-scale systems simulating different AO configurations in closed loop.

Phase 5: (12 months) Real-time validation on a test bench. Depending on the success of the laboratory experiments, the multi-core AO system can also be tested on-site e.g. at the William Herschel Telescope on La Palma.

Phase 6: (3 months) The PHD student writes her thesis inlcuding all the results of the project which will be defended in front of the Promotion Committee. The individual results will be presented at international conferences and in scientific journals, for which time is planned in each work package.



Status & Results

Phase 1 and 2 are finalized. A wavefront reconstruction algorithm which provides an extension of SABRE for intensity measurements was developed and published.  Advances on the distributed extenstion of SABRE haven been made in close collaboration with Coen de Visser of the Aerospace Engineering Faculty at TUD.  Currently  the method is being implemented for the AO end-to-end simulator YAO to evaluate its closed-loop performance in preparation for a real-time implementation. Collaborations with researchers from the teams working on the GMT and TMT have been initiated providing data and a simulator for the respective planned large-scale telescopes. Three master students haven been/are working on the project.



Involved Partners

The work is mainly done by Delft Center for Systems and Control, in colaboration with staff members of other facultys and a potential role for TNO.


Recent Publications:

J. Silva, E. Brunner, A. Polo, C. de Visser, M. Verhaegen, “Wavefront Reconstruction Using Intensity Measurements for Real-time Adaptive Optics”, European Control Conference, 2014.


E. Brunner, C. de Visser, J. Silva, M. Verhaegen, “Distributed Wavefront Reconstruction with SABRE for real-time Large Scale Adaptive Optics Control”, SPIE Atronomical Telescopes & Instrumentation, 2014.


E. Brunner, J. H. Girard, “Atmospheric parameter estimation from AO wavefront sensing data: Application of the FADE method with NACO”, SPIE Atronomical Telescopes & Instrumentation, 2014.


E. Brunner, J. Silva, C. de Visser, M. Verhaegen, “Compressive Sampling in Intensity Based Control for Adaptive Optics”, World Congress of the International Federation of Automatic Control, 2014.

 



Glossarium

SABRE
Spline based ABerration Reconstruction for SH slope measurements, wavefront reconstruction algorithm


D - SABRE
Distributed Spline based ABerration REconstruction, wavefront reconstruction algorithm


SABRE - I
Spline based ABerration Reconstruction for SH intensity measurements, wavefront reconstruction algorithm



Powered by VIADOMUS