Interferometric Optical System Design and Signal Processing Technology for Speckle Noise Reduction

Abstract

Abstract

Recently, as demands for precision and speed in mechanical systems have recently increased, it has become crucial to identify vibrations and noise within these systems. Interferometry-based sensors can precisely and accurately measure physical quantities such as distance, velocity, vibration, and shape by creating interference between a laser reference beam and a received beam reflected from an object. Thanks to these advantages, demand for interferometric sensors is surging across various industries, including machinery, precision measurement, automotive, robotics, semiconductors, and defense. While interferometric sensors enable precise and accurate measurements, the intensity of the received beam incident on the detector varies depending on the surface roughness of the scanned or vibrating object, resulting in interference signals with different amplitudes. In “necking” regions, where the amplitude of the generated interference signal is smaller than that of a specific signal, making frequency extraction difficult, spike-shaped speckle noise occurs in the interferometer’s velocity or distance signal. This phenomenon degrades the signal-to-noise ratio （SNR）, which consequently reduces the accuracy of the interferometer’s velocity and distance measurements. Speckle noise reduction technology that utilizes optical systems equipped with multiple photodetectors, deep learning, and signal post-processing can reduce speckle noise more effectively than conventional methods using a single detector. However, various issues–such as high computational load and hardware dependency–impose limitations on real-time speckle noise reduction. To overcome these limitations, this study employs a quadrant photodiode （QPD） in the receiver of a conventional heterodyne interferometer optical system instead of a single detector to obtain spatially separated interference beams. In this paper, we propose and compare the “amplitude-proportional weighted average method,” which applies amplitude-based weights to generate a single signal composed primarily of good signals, and the “log-weighted average method,” which applies gain to regions with weak interference signals to form a single signal while preserving the “necking” regions where signal is lost as much as possible. The fabricated LDV was used to measure the velocity signals of objects and to determine the speckle noise occurrence rate. The experiments were conducted in environments where the speckle noise rate of the velocity signals obtained from a single detector was 20–35% and 35–50%. The results showed that the proposed amplitude-proportional weighted averaging method reduced speckle noise by an average of up to 70.5%, while the log-weighted average method achieved an average speckle noise reduction rate of up to 66.0%. Consequently, this paper presents a real-time signal processing solution that improves the measurement reliability and commercial viability of interferometer systems, demonstrating excellent potential for the high-speed real-time frequency measurement industry.