Random matrices exhibit interesting statistical properties which are studied under random matrix theory (RMT). In this research study, we present a novel approach for fiber optic distributed acoustic vibration sensing (DAS) systems which is based on the recent results of RMT. Our focus is the phase-sensitive optical time domain reflectometry (φ−OTDR) systems and the evaluation of the RMT at the photo-detection output. Inspired by the successful application of RMT in diverse signal processing applications, the RMT based signal detection methodology is transferred to DAS domain. The classical spectral theorem is revisited with special emphasis on the covariance of the measured Rayleigh backscattered optical energy which is a Wishart type random matrix. A real φ−OTDR system is evaluated for experimental verification of the statistical distributions of the extreme eigenvalues of the optical covariance matrix. It is shown that even with limited measured data, after proper conditioning and scaling of the optical detector output, the empirical bulk eigenvalue distributions are in good agreement with the analytical proof for the infinite data assumption. It is experimentally verified that the extreme eigenvalues of the optical covariance are bounded by the Marchenko-Pastur theorem and any outlier can be considered as a vibration presence. Additionally, it is shown that the eigenvalue bounds can be used to detect and track the vibrations along a fiber optic cable route.
Distributed acoustic sensing (DAS) based on phase-sensitive optical time domain reflectometry (OTDR) is being widely used in several applications and attracting significant research interest. The main challenge in coherent detection-based phase-sensitive OTDR systems is the speckle-like background noise which impacts the detection performance and conventional techniques are not suitable for detecting weak vibrations under strong background noise. Recently, we proposed a temporal adaptive filtering (AMF) technique to reduce the background noise in phase-sensitive OTDR systems. The AMF method is based on linear filtering of the optical backscattered signals and the filter coefficients are computed from the observed data. In this study, after briefly reviewing the fundamental theory underlying the adaptive algorithm, we present the effectiveness and performance results of the AMF technique with the field tests. The impact of the diagonal loading level which is used to solve the ill-conditioning of the estimated noise covariance matrix is investigated. Performance dependence of the AMF technique on filter size and a comparison with the conventional trace averaging is presented. It is demonstrated that with the AMF technique, more than 10 dB of SNR values can be achieved without introducing additional optical amplifier stages in the DAS hardware. It is shown that intruder activities 25 m far away from a buried SMF-28 fiber underground can be detected with the proposed technique efficiently.
Proc. SPIE. 9852, Fiber Optic Sensors and Applications XIII
KEYWORDS: Signal processing, Acoustics, Fiber optics sensors, Digital filtering, Filtering (signal processing), Fiber optics, Signal detection, Signal to noise ratio, Electronic filtering, Monte Carlo methods
We introduce a new approach for distributed fiber optic sensing based on adaptive processing of phase sensitive optical time domain reflectometry (Φ-OTDR) signals. Instead of conventional methods which utilizes frame averaging of detected signal traces, our adaptive algorithm senses a set of noise parameters to enhance the signal-to-noise ratio (SNR) for improved detection performance. This data set is called the secondary data set from which a weight vector for the detection of a signal is computed. The signal presence is sought in the primary data set. This adaptive technique can be used for vibration detection of health monitoring of various civil structures as well as any other dynamic monitoring requirements such as pipeline and perimeter security applications.