Abstract
Optical interferometers as non-contact measurement devices are very desirable for the measurement of surface roughness and
topography. Compared to phase shifting interferometers (PSIs) with a limited measurement range and a scan step of maximum
λ/4, the optical interferometers like low coherence interferometers (LCIs) evaluating the degree of fringe
coherence allow a larger vertical measurement range. Their vertical measurement range is only limited by the scan length
allowed by the linear piezo stage and the coherence length of the light source. To evaluate the obtained data for a large
range, the common LCIs require much computation time. To overcome this drawback, we present an evaluation algorithm based
on the Hilbert-Transform and curve fitting (Levenberg–Marquardt algorithm) using Compute Unified Device Architecture
(CUDA) technology, which allows parallel and independent data evaluation on General Purpose Graphics Processing Unit
(GPGPU). Firstly, the evaluation algorithm is implemented and tested on an in-house developed LCI, which is based on
Michelson configurations. Furthermore, we focus on the performance optimization of the GPU-based program using the
different approaches to further achieve efficient and accurate massive parallel computing. Finally, the performance
comparison for evaluating measurement data using different approaches is discussed in this paper.