GPU Acceleration in Wave-Front Sensorless Adaptive Wave-Front Correction System

As for the wave-front sensorless adaptive wave-front correction system based on the stochastic parallel gradient descent (SPGD) algorithm, its convergence speed is too slow to satisfy the real-time requirement of a wireless optical coherent communication system. The parallel processing base on the SPGD algorithm is introduced and the graphics processing unit (GPU) parallel computing is used to improve the convergence speed of the correction system. The average gray value of the surrounding 400 pixels centered on the centroid of the real-time spot detected by CCD camera is employed as the value of system performance index. GPU multithreading operation is used to accelerate the solving process of the performance index and the updating process of the deformable mirror control voltage vector. The results from the indoor experiments and the external coherent light experiments show that the Strchl ratio is larger than 0.8 and the maximum time acceleration ratio is up to 8.6. Moreover, the convergence speed of the GPU accelerated wave-front correction system is improved and simultaneously the correction effect is ensured.