Robust Signal Extraction Based on Time-Frequency Joint Analysis and GRNN for a Laser SMI System

In pursuit of vibration recovery, with subwavelength precision based on laser self-mixing interferometry (SMI), the estimation of two key parameters, i.e., the optical feedback parameter
$C$
and linewidth enhancement factor
$\alpha$
, is of vital importance. However, the above process is not ideal in terms of accuracy and real-time performance. Aiming to overcome these deficiencies, a time-domain-based frequency-domain processing method (TFPM) has been proposed for the SMI system in this paper, which realizes robust and fast extraction of the vibration signals without the estimation of the parameters and does not require fringe detection. The method achieves the extraction of motion information via a mathematical transformation, namely, short-time Fourier transform (STFT), in which the limitations caused by STFT are alleviated by using generalized regression neural network (GRNN). In processing the SMI signal of the target with non-stationary motion, the TFPM finally achieves an expected root-mean-square (RMS) error of 0.24
$\%$
, which demonstrates the high performance of the proposed method.