Berlin 2012 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 18: Posters: Statistical Physics in Biological Systems
BP 18.33: Poster
Mittwoch, 28. März 2012, 17:30–19:30, Poster A
Fast GPU-based Hologram Generation for Real-Time Optical Neurostimulation — •Johannes Hagemann1, Hecke Schrobsdorff2, and Stephan Kramer3 — 1Institut f. Röntgenphysik, Friedrich-Hund-Platz 1, 37077 Göttingen — 2MPI f. Dynamik und Selbstorganisation, Institute for Nonlinear Dynamics, Am Fassberg 17, 37077 Göttingen — 3Institut f. Numerische und Angewandte Mathematik, Lotzestr. 16-18, 37083 Göttingen
For optical neurostimulation, holograms have proven to be particularly efficient. Fast and flexible stimulation is crucial for e.g. controlled synaptic learning experiments. Since a spatial light modulator used for hologram generation can only shift phases, hologram computation is a complex process. Up to now, no direct method to determine a corresponding phase mask exists. Current algorithms consist of a series of iterative fourier transforms and projections, with sometimes very slow convergence speeds. Fortunately, the matrix formulation of existing algorithms is data-parallel and thus can directly be implemented on a GPU. We present a CUDA-based implementation for computing phase-only holograms which are used for a selective excitation of individual light-sensitive neurons. We focussed on a modular structure to be flexible for changing experimental hardware or for incorporating advancements in parallel computing or new algorithms. We compare the performance of the well-known GS algorithm [1] with the recently introduced Relaxed Averaged Alternating Reflections [2].
[1] R. W. Gerchberg and W. O. Saxton, Optik 35, 237 (1972)
[2] D. R. Luke, Inverse Problems 21,37 (2005)