Signal processing for brillouin distributed optical fiber sensing systems

Abstract

Brillouin distributed optical fiber sensors have been extensively studied over the past few decades. Among them, stimulated Brillouin scattering (SBS) based Brillouin optical time domain analysis (BOTDA) sensors have experienced rapid development and attracted ample attention in fundamental scientific research as well as in practical applications due to their ability regarding distributed temperature and strain measurements over long sensing fibers with good accuracy and spatial resolution. The use of BOTDA sensors in many practical applications requires better accuracy over longer sensing distances within a shorter signal processing time than can be provided using current techniques. The performance improvement of signal processing techniques is thus vital for expanding the use of BOTDA sensors in real-world applications. In this thesis, the limitations of widely-used signal processing techniques for BOTDA sensors are identified and analyzed. To overcome the limitations of existing methods, several signal processing techniques are proposed and demonstrated experimentally. The use of artificial neural network (ANN) is proposed and demonstrated to extract temperature distributions directly from Brillouin gain spectrum (BGS) measured by BOTDA sensors. No curve fitting process is needed, and the process of determination of Brillouin frequency shift (BFS) and conversion from BFS to temperature are also not required. Moreover, ANN can be trained and optimized using input-output patterns obtained either from BOTDA experiments or through appropriate theoretical modeling before it is applied in temperature extraction applications. The results validate that the performance of BOTDA sensors with ANN for temperature extraction does not degrade significantly at large frequency steps. With the performance comparison between ANN and other techniques, ANN exhibits better accuracy, greater tolerance to measurement error and significantly faster signal processing speed, especially when the BGSs are acquired by adopting large frequency steps. Thus, ANN can be a potential tool to accurately extract temperature/strain distributions for fast monitoring applications of future BOTDA sensors. Besides ANN, the use of the principal component analysis (PCA) based pattern recognition method is also proposed and demonstrated experimentally to extract temperature distributions from the BOTDA-measured BGSs along the fiber. In the proposed technique, a reference database with relevant theoretical BGSs of known characteristics are constructed where the BGSs are processed using PCA to extract very few but the most significant feature vectors, i.e., limited number of principal components to represent each BGS. On the other hand, the measured BGSs obtained from BOTDA measurements are also processed using PCA to extract the reduced-size feature vectors and then for each given feature vector, the best match in the reference database is determined based on statistical distance measurement. The results show that the technique can provide better accuracy than other techniques for extracting temperature distributions along the fiber. Moreover, the computational complexity of the matching process with reduced-size feature vectors obtained through PCA decreases significantly, thereby making the system suitable for fast monitoring applications. The signal processing speed of this technique can also be customized to make it suitable for applications with special timing requirements. Therefore, the proposed PCA-based technique can be an attractive alternative for extracting temperature distributions along the fibers in BOTDA sensors. The accuracy of BOTDA sensors is ultimately determined by the signal-to-noise ratio (SNR) of the measurements. A common and widely-used way to improve such SNR is to adopt a large number of BOTDA trace-averaging, which can be very time-consuming, especially for a long sensing fiber. One of the possible solutions is to acquire the BGSs using a small number of trace-averaging to reduce the BOTDA measurement time and, then to use a faster denoising algorithm to enhance the SNR. To study the feasibility of using such denoising algorithm, wavelet transform-based signal denoising technique is demonstrated experimentally to improve the measurement SNR of the BOTDA-measured signals. The results show that such technique improves the measurement SNR by several orders of magnitude and hence, measurement accuracy can be improved significantly without sacrificing the spatial resolution. The signal processing time of existing methods is quite long. Moreover, the use of the denoising algorithm also extends the processing time in the signal processing stage. To take full advantage of the denoising algorithm for faster and accurate signal processing, wavelet denoising technique (WDT) based ANN is proposed and demonstrated in the thesis. The results confirm that this WDT-based ANN significantly improves the measurement accuracy as well as signal processing speed to make BOTDA sensors more suitable for faster operation. The development of more efficient and faster signal processing techniques over existing methods in the thesis will be a leap forward for the expansion of BOTDA sensors in diverse real-world applications.