Real Time control

How to handle the extreme AO demands

Click here for Extended Information


Identify a suitable real-time control architecture that can handle the EPICS-AO demands. The expected real-time performance will be validated on a laboratory setup.


The project phases can be formulated as:

A. Exploration of current available architectures and examination of the suitability for the

EPICS-AO application taking into account processing power, flexibility, scalability and maintenance aspects.

B. Selection of the most promising architecture and integration into an existing adaptive optics breadboard. Optimization of the performance to fully utilize the hardware capabilities.

C. Analysis of the experimental results and extrapolation to the EPICS-AO case in order to predict the real-time performance.

D. Document the results and provide input of the real-time computing platform to the ESFRI research on “Fast and hgh performance control”.


The project showed thateven with present-day hardware the demands of EPICS AO can be met.

Involved Partners
TNO and Delft Center for Systems and Control.

Progress, Finished

This project is related to the “Fast and High Performance Control” project. Where that project focuses on the software side, this project looks at the hardware architecture. The main project goal is to identify a suitable real-time control architecture that can handle the EPICS-AO demands. The expected real-time performance will be validated on a laboratory setup.

The project can be split up into 4 work-packages in total.

WP 1. Exploration and assessment of hardware architectures

Several real-time hardware architecture options will be assessed. The essential criteria to consider are:

• Processing power, data-throughput bandwidth

• Real-time capabilities

• Programming flexibility

• User flexibility

• Scalability

• Maintenance (hardware availability, support, forward compatibility with 2020+ hardware)

• Hardware costs

Possible hardware architectures, as currently foreseen, are the following:

1. A Graphics Processing Unit (GPU) based platform consisting of multiple NVIDIA CUDA, FERMI, ATI Stream or OPENCL supported cards.

2. A multi- or many-core CPU based system utilizing the distributing capacities of new generation CPU’s (e.g. with OpenMP).

3. A (DSP implementation on a) Field-programmable gate array (FPGA) on a dedicated board or specifically integrated on the framegrabber of the wavefront sensor.


WP 2. Implementation and code optimization

Both the classical AO control algorithm and a higher order variant will be implemented on the selected platform. The first iteration in this process is a straightforward software implementation. This will not give the best real-time performance. Subsequently, the real-time code will be optimized in order to fully exploit the capabilities of the architecture. Code optimization for the architecture will be the main activity of this work-package.


WP 3. Validation and demonstration of performance

The hardware and developed control software will be integrated into an experimental set-up, consisting of a wavefront sensor, a deformable mirror and the real-time control system. The experiment is focused on the validation of the expected real-time behaviour of the control architecture only, rather than mimicking a scaled-down EPICS AO case. The experiment will reveal whether the control architecture indeed meets the requirements, in particular processing power, data-throughput bandwidth, real-time capabilities, programming flexibility, user flexibility and scalability.


WP 4. Full-scale analysis and conclusions

The results of the previous WPs will be extrapolated to EPICS-AO dimensions. In this way an estimation of the performance of the tested hardware architecture for the EPICS-AO case can be given.


The project has been finished and the results have been presented to ESO. Based on this research it is concluded that 4 GPU’s should be sufficient to control the M4 of the E-ELT, and therefore that even with present day technology this can be solved.

Involved Partners

The project was done by  TNO and Delft Center for Systems and Control.


Real time control
Controlling aspects of an instrument (in this case a deformable mirror) during the measurements with live input of the changes. For deformable mirrors that want to correct for atmospheric disturbances the input and therefore the calculation of the required changes should happen about 1000 times per second.

Powered by VIADOMUS