Multi-source Separation Method Of Distributed Optical Fiber Vibration Sensing Signal Under Two "Blind" Conditions

Fiber-optic distributed acoustic sensors(DAS)based on the phase-sensitive optical time-domain reflection(?-OTDR)principle,using existing communication cables as a medium,play an important role in urban infrastructure monitoring and natural disaster prediction.Significant progress has been made in single-source identification of DAS,but independent of multi-source separation,the intelligent processing capability and algorithmic robustness of highly sensitive DAS systems in complex urban interference environments can not yet meet the actual application requirements,especially in the presence of multiple interference from different sources on the ground and complex buried conditions,when multiple unavoidable and unpredictable interference sources coexist and are superimposed on the same fiber optic receiving point,the accurate detection and identification of DAS sensing target signals face huge challenges and become the biggest technical bottleneck limiting its scale application in urban security awareness and warning.Therefore,in this paper,we propose the first blind separation method for DAS multi-source mixed signals under "double-blind" conditions to address the challenges of unknown sudden interference sources and unknown buried conditions.The specific work done in this paper is as follows.(1)Under the "double-blind" problem of unknown superimposed source number and mixing mode,a feasible blind source number estimation method is proposed to estimate the mixing source number by the singular value distribution of the signal subspace.Based on the propagation mechanism of a single vibration source at the surface or near the surface,the multi-source mixing mechanism of the superimposed vibration sources received by the DAS array is discussed based on the typical geological conditions of uniform and non-uniform,and the assumptions from linear transient mixing model to linear convolutional mixing model are made to continuously approximate the real mixing propagation mode under complex burial conditions.(2)A multi-source separation method under the assumption of linear transient mixing model is proposed-the DAS multi-source mixing separation method based on the improved Fast ICA.The non-Gaussianity of the signal is the basic criterion and basis of blind separation,and the basic idea of blind separation is to non-Gaussianize the joint distribution of the signal.The separability of the blind multi-source separation method of DAS is analyzed according to this criterion,and the feasibility of the separation is demonstrated experimentally.The separation performance is also improved by using wavelet soft threshold denoising method.(3)The separation experiments are carried out by three experimental scenarios of simulated mixing,real mixing under laboratory conditions,and real mixing under outdoor ground burial conditions to verify the effectiveness of the algorithm.The results show that the proposed algorithm can successfully separate the DAS multi-objective mixed vibration signals under the assumption of linear transient mixing model.In addition,the reasons for incomplete separation and the evaluation methods of the simulation and actual separation performance are discussed.