Experiment Design for Subspace Identification of LTI/Wiener Systems for large scale distributed systems
|Project members:||prof.dr.ir. M. Verhaegen (Michel), PhD position - Vacancy|
|Keywords:||System identification, Spatial-temporal large-scale systems, Control of high-resolution imaging, Optics and imaging, Machine learning|
|Sponsored by:||European Research Council|
|Principal Investigator:||Prof. Michel Verhaegen|
(Team Leader of the Numerics for Control and Identification Group)
For the restoration of images knowledge of the spatial-temporal dynamics of the aberrations is crucial. In the new context of the iCON Advanced Grant Research Project sponsored by the European Research Council, image restoration will be addressed within the context of real-time feedback control, making use of actuators like deformable lenses or mirrors in addition to classically used image sensors.
Within the context of a new data driven approach based on Subspace Identification to acquire these aberration dynamics the design of experiments in real-time open or closed will be investigated in the scope of this project.
The challenges for the experiment design in the scope of Subspace Identification methods are twofold:
- The 2D spatial and temporal dynamics: The requirement of high resolution imaging generally leads to the use of a large number of actuators and sensors to attain the necessary spatial resolution. Considering these numbers in the order of 104 with a temporal data acquisition in the order of a few kHz, excludes the use of present centralized subspace identification methods. Instead a distributed methodology allowing execution on a multi-core GPU/GPU where each core has access to a small subset of sensors and actuators only. In addition for data reduction the algorithms will be integrated with compressive sensing.
- Inferential sensing: The quantity of interest, i.e. the wavefront aberration or the phase of the optical wave, cannot be measured directly. It has to be reconstructed from intermediate signals, like the wavefront spatial slope measurements given by a Shack-Hartmann sensor or the readings of the CCD science camera. A data driven approach will be developed for the calibration of the relation between the measured quantities and the image restoration metric and how the experiments should be tuned to identify the spatial-temporal models that enable to restore the images optimally wrt the selected metric.
The new algorithmic developments will be demonstrated in 2 state of the art laboratory demonstrators that will be built up in the Smart Optics Laboratory of Prof. Michel Verhaegen. The first demonstrator is a breadboard emulating the large dimensionality of the Adaptive Optics control problem in the European Extreme Large Telescope and the second is about multi-photon microscopy.
iCON is sponsored by the Advanced Grant Program of the European Research Council. This funding will bring a core team of 6 temporary researchers together with world wide leading experts for a period of 5 years starting early 2014.