Sparse representation-based noise reduction for BOTDR monitoring signals in foundation pit anchor cables

Abstract Brillouin optical time-domain reflectometer (BOTDR) systems are commonly challenged by low signal-to-noise ratio (SNR) in foundation pit monitoring. This study proposes a noise-suppression method for BOTDR signals utilizing sparse representation (SR). The method involves creating an initial dictionary from the eigenvectors of the normalized graph Laplace matrix. The K-singular value decomposition and orthogonal matching pursuit algorithms are combined to update the dictionary and coefficient matrix, facilitating the SR of the signal’s intrinsic features and the removal of random noise. This results in improved quality of the reconstructed signal. An experimental system for BOTDR temperature sensing was developed to assess the algorithm’s denoising capabilities. The algorithm showed significant improvements in SNR and reductions in sample entropy (SE) compared to techniques such as wavelet threshold denoising, empirical wavelet transform, and empirical mode decomposition. Specifically, the average SNR increase was 27.4%, 15.4%, 13.1%, and 17.9%, while the average SE decrease was 24.4%, 16.0%, 15.4%, and 47.9% for the respective comparisons. The proposed Laplace-based dictionary also outperformed the discrete cosine transform dictionary. Field tests were conducted at the Beijing Stomatological Hospital, where the algorithm applied to on-site anchor cable fiber optic monitoring signals, achieving an average SE decrease of 45.2%. The research provides an effective denoising method for the application of BOTDR technology in foundation pit monitoring, underscoring the approach’s novelty and practical application.