| We apply phase retrieval, a method of wavefront sensing employing intensity measurements and a simple experimental arrangement, to problems in optical metrology.;The limits of the technique, in terms of convergence of the beam being tested and the amount of wavefront error that can be measured are defined theoretically using geometrical optics and sampling theory. These limits suggest methods for expanding the range over which phase retrieval is capable.;We explore the precision and accuracy of phase retrieval through numerical experiments using realistic simulated data. The results show that in the presence of realistic error and noise sources, phase retrieval can retrieve wavefronts with errors on the order of lambda/1000 RMS, sufficient for many applications in optical metrology. We also show methods for mitigating certain errors by modeling non-ideal effects in the phase retrieval algorithm.;We have conducted several experiments, measuring optical surfaces and transmitted wavefronts. Experiments demonstrated the repeatability of measurements, using disjoint raw data, on the order of lambda/1000 RMS. Experimental results were compared to a ray trace model and measurements made using a Shack-Hartmann wavefront sensor. In both cases, agreement was on the order of lambda/100 RMS.;We described and demonstrated, in simulation, a new phase retrieval algorithm that solves for both the amplitude and phase in the pupil plane. This approach mitigates reconstruction errors due to uncertainty about the pupil amplitude distribution.;We have developed and demonstrated experimentally a new phase retrieval algorithm that uses multiple transversely translated intensity measurements that allows us to measure the wavefront of highly convergent beams that are not possible to measure using conventional phase retrieval. Reconstructions using the method agree to about lambda/100 RMS.;The work described here demonstrates that phase retrieval is a viable and realistic method for optical surface and wavefront sensing. It describes the limitations of the method and possible ways to expand them. We explore some of these methods through algorithmic improvements. |
