Complexity Analysis and Parallel Implementation of the Well Optimized Linear Finder (WOLF) Atmospheric Turbulence Compensation (ATC) Algorithm on GPU Technology with Comparative Analysis of Traditional Phase Diverse ATC Methods
Abstract
A new high-speed, high-accuracy general transfer function estimation method for diversitybased problems, known as the Well Optimized Linear Finder Algorithm (WOLF Algorithm), has been developed [1]. Compared to the traditional phase diversity ATC
methods, the WOLF Algorithm has less computational complexity and better approximation of the Optical Transfer Function (OTF) and consequently the entrance pupil
phases. We compare the time complexity between the traditional Phase Diversity atmospheric turbulence compensation approach and the WOLF Algorithm. The results
show that the WOLF Algorithm is dramatically faster than traditional phase diversity
methods theoretically. We also implement the time complexity results of the WOLF
methodology using serial processing on a Central Processing Unit (CPU) architecture
(2.8 GHz Quad-Core Intel Core i7 processor). The results show that the simulated run
time for the WOLF methodology with a 256 × 256 pixel image is 8.2 seconds which
matches our theoretical results within one standard deviation. Finally, we simulate the
WOLF methodology time complexity using parallel processing on several GPUs. The
results for the WOLF methodology with a 256 × 256 pixel image on the fastest GPU
processor (Nvidia Tesla T100) was 3.7740 × 10−3
seconds (264.97 Hz), which is 1224.05
times faster than the implementations on a CPU architecture.