Optical testing, having the merits of non-destruction and high sensitivity, provides a vital guideline for optical manufacturing. But the testing process is often computationally intensive and expensive, usually up to a few seconds, which is sufferable for dynamic testing. In this paper, a GPU-accelerated phase extraction algorithm is proposed, which is based on the advanced iterative algorithm. The accelerated algorithm can extract the right phase-distribution from thirteen 1024x1024 fringe patterns with arbitrary phase shifts in 233 milliseconds on average using NVIDIA Quadro 4000 graphic card, which achieved a 12.7x speedup ratio than the same algorithm executed on CPU and 6.6x speedup ratio than that on Matlab using DWANING W5801 workstation. The performance improvement can fulfill the demand of computational accuracy and real-time application.
Generally, in order to gain high accuracy in aspheric testing, a piece of high-quality CGH (computer generated hologram) is inserted behind transmission sphere to generate specified wave-front to match aspheric part. According to the difference in function, the CGH is divided into 2 parts: the center region, called as testing hologram, is used to generate specified aspheric wave-front; the outer ring, called as alignment hologram, is used to align the location of CGH behind transmission sphere. Although alignment hologram is used, there is still some adjustment error from both CGH and aspheric part, such as tilt, eccentricity and defocus. Here we will stimulate the effect of these error sources on the accuracy that is rms after the piston, tilt and power are removed, when testing a specified aspheric part. It is easy to conclude that the total measurement error is about 2 nm and the defocus of CGH contributes most.
Comparative simulations are designed to evaluate the noise suppression performance of three typical phase shifting algorithms under different testing environments. The results show that random phase shifting algorithm is robust under different level of noise and has higher testing accuracy than conventional normal steps phase shitting algorithm while it’s computationally low. Guidelines are given to choose a proper PSI algorithm under a certain testing environment with noises.