The distributed optical fiber sensing technology based on Brillouin scattering faces several challenges in accelerating its engineering and industrialization, with the signal to noise ratio (SNR) being a critical metric that directly determines the system’s sensing range and detection accuracy. Although Brillouin optical time-domain reflectometry (BOTDR) exhibits high sensitivity in strain and temperature detection, the inherently weak optical power of spontaneous Brillouin scattering (SpBS) results in a low system SNR, which limits its performance in long-distance applications. Traditional SNR enhancement methods, such as optical pulse coding, bidirectional Raman amplification, multi-wavelength detection, and differential detection, are effective in improving SNR but often introduce increased system complexity and hardware costs. With the rapid advancement of digital signal processing (DSP) technologies, enhancing SNR through digital approaches not only improves system performance but also significantly reduces hardware expenses and enables more flexible and efficient technological iterations. Current BOTDR systems primarily employ two time-frequency analysis methods: frequency-swept (FS) and short-time Fourier transform (STFT) techniques. Due to the distinct mechanisms of acquiring time-frequency data in STFT-BOTDR compared to FS-BOTDR, directly applying conventional filtering and denoising methods to the acquired broadband signals may result in the loss of critical frequency-domain information, thereby degrading system performance. To address this limitation, we propose a novel denoising strategy tailored for STFT-BOTDR, integrating a variational mode decomposition (VMD) and triangular topology aggregation (TTA) optimized adaptive denoising algorithm to enhance the SNR of demodulated data. This method requires no additional hardware components, offering existing STFT-BOTDR systems an adaptive and efficient denoising solution. The proposed approach improves instrument performance, enhances cost-effectiveness, and strengthens engineering practicality, thereby providing a viable pathway for advancing distributed fiber sensing technologies in industrial applications.
We propose an adaptive denoising strategy integrating VMD and TTA algorithms for the signal processing mechanism of STFT-BOTDR systems. First, the acquired broadband signals are segmented via a sliding window with a fixed length and appropriate step size. For each segment, a Fourier transform is performed to extract the Brillouin gain spectrum (BGS), and the BGS sequences are aligned chronologically to construct a time-frequency distribution curve. Subsequently, the TTA algorithm is employed to optimize the combination of VMD decomposition parameters (i.e., the number of modes and penalty factor) for time-domain signals corresponding to each frequency point in the time-frequency curve, with the signal kurtosis value serving as the fitness function. Specifically, the TTA algorithm constructs multiple similar triangular topology units and iteratively optimizes them through generalized and local aggregation strategies to identify the optimal VMD parameters that maximize denoising efficacy, thereby enhancing adaptability and precision. Following parameter optimization, VMD is applied to decompose each time-domain signal at the same frequency into multiple intrinsic mode functions (IMFs). To further refine denoising performance, a joint criterion based on sample entropy and variance contribution rate is introduced to quantify the signal reconstruction range. The denoised signal is then obtained by reconstructing the retained IMFs. The proposed algorithm is comprehensively compared with discrete wavelet transform (DWT) and complementary ensemble empirical mode decomposition (CEEMD) in terms of SNR improvement, temperature measurement accuracy, denoising effectiveness under varying noise levels, and effect on spatial resolution. Experimental results validate the superiority of the proposed approach in balancing noise suppression, signal fidelity, and system adaptability.
Through comparative analysis of various denoising algorithms, we validate the superiority of the TTA-VMD algorithm in denoising fiber-optic line data. After TTA-VMD denoising, the “spikes” in most fiber-optic data are effectively suppressed, which results in significantly smoother denoised data. Compared to the other two denoising algorithms, TTA-VMD demonstrates a notable advantage in improving SNR. Specifically, the SNRs of raw data (RAW), DWT, CEEMD, and TTA-VMD are 33.62 dB, 39.65 dB, 39.13 dB, and 43.53 dB, respectively, with TTA-VMD achieving a 9.91 dB improvement over raw data (Fig. 10). Additionally, after TTA-VMD denoising, the fluctuation of the Brillouin frequency shift (BFS) is significantly reduced, with its standard deviation decreasing from 2.40 MHz (raw data) to 1.09 MHz, which indicates substantial improvement (Fig. 11). Under pulse widths of 50, 60, 70, 80, 90, and 100 ns, the TTA-VMD algorithm achieves an average SNR improvement of 9.08 dB across all temperature intervals. In contrast, the SNR improvements of DWT and CEEMD algorithms do not exceed 6.00 dB. This demonstrates the superior performance of TTA-VMD in SNR enhancement, effectively reducing noise and enhancing signal quality. Further analysis reveals that the standard deviation of BFS after TTA-VMD processing decreases to approximately 1 MHz, outperforming other algorithms such as DWT and CEEMD (Fig. 12). Even under low SNR conditions, TTA-VMD still effectively improves both SNR and BFS standard deviation (Fig. 14). Moreover, the TTA-VMD algorithm exhibits significant improvement in the smoothness of 2D BGS distribution images, particularly in the identification of heated sections, where denoised results show enhanced clarity and noise suppression (Fig. 15). When temperatures are varied (25, 30, 35, 40, 45, and 50 ℃) and 50peMD demonstrates clear advantages over DWT and CEEMD in terms of smoothness and noise suppression in heated sections. The coefficient of determination (R2) for noise suppression in heated section further confirms the superiority of TTA-VMD, which reaches as high as 0.99011, significantly higher than that of DWT and CEEMD (Fig. 16). Regarding temperature measurement accuracy, TTA-VMD exhibits the smallest measurement deviations and the highest stability across all temperature intervals, while DWT and CEEMD show smaller errors at low temperatures but significantly larger errors at high temperatures (Table 1). In terms of applicability, TTA-VMD demonstrates robust denoising capabilities across various fiber types and lengths, with stable performance and adaptability (Fig. 18). For spatial resolution, the proposed algorithm exhibits some dependency: narrow pulses probing short temperature-varying regions slightly degrade spatial resolution, but this effect diminishes as the pulse width and heating length increase (Fig. 19).
We propose a new method for denoising time-frequency data based on STFT processing of time-domain curves at the same frequency point, which achieves overall denoising of time-frequency data and effectively avoids the problem of data degradation caused by directly filtering and denoising the original signal. Therefore, we combine VMD and TTA algorithms to propose an adaptive denoising algorithm. This algorithm adaptively optimizes the parameter combination of VMD through TTA, solving the problem of insufficient decomposition and mode mixing that may be caused by empirical parameter settings in traditional VMD methods. The use of VMD for signal decomposition and reconstruction, along with the quantification of signal reconstruction range through the joint criterion of sample entropy and variance contribution rate, further improves the denoising effect. The experimental results show that the proposed adaptive denoising algorithm can effectively improve the SNR of the signal on the sensing fiber while maintaining high spatial resolution. Under the sensing fiber length of 0.8 to 5 km, the SNR is improved by at least 8 dB, and the BFS standard deviation at the end of the fiber is reduced to about 1 MHz. Compared with other denoising algorithms, the TTA-VMD algorithm exhibits the smallest measurement deviation and the highest temperature measurement stability across different temperature ranges, which further demonstrates its advantages in practical applications. The algorithm in this study has a dependence on the system detection pulse width and the length of the temperature change area. Using narrow pulses to detect short temperature change areas slightly affects spatial resolution, but as the detection pulse width and heating length increase, their effect gradually decreases. In addition, this method only requires post-processing of the collected signals, without additional hardware support, and avoids the increase in additional costs while improving signal processing performance. Therefore, the TTA-VMD adaptive denoising algorithm proposed in this article provides an efficient and adaptive denoising scheme for existing BOTDR systems based on STFT, thus significantly improving system performance, cost-effectiveness, and practicality for engineering applications.