Performance Analysis Of BFS Extraction Technology In BOTDR Based On Different Fitting Algorithms

In recent years,with the development of optical communication industry,optical fiber sensor has become an important part of structural health monitoring field,and has been widely used in large civil engineering,petroleum and petrochemical,tunnel traffic,high voltage transmission lines and other fields.With the continuous development of distributed optical fiber sensors,it is difficult to meet the real-time requirements of more and more systems with the exponential growth of distributed optical fiber sensor information by using the traditional Lorentz fitting method.How to extract sensor information quickly and accurately is one of the problems to realize real-time sensing.In order to solve the problem of temperature and strain cross sensitivity in Brillouin distributed sensing system,the distributed sensing system based on multi-peak Brillouin scattering emerged at the right moment,and the extraction of multi-peak Brillouin frequency shift information has become one of the main research directions.Based on the analysis and summary of traditional BFS extraction technologies,this paper proposes a scheme to extract Brillouin sensor information using CRP and BP neural network.The specific work and innovations are as follows:(1)The Brillouin optical time domain reflectometry system(BOTDR)was built by using the optical domain frequency conversion coherent detection technology.The Brillouin gain spectrum in single-mode fiber under experimental conditions was determined as the infinite approaching Lorentz type,which was taken as the reference spectrum type for the following simulation.Several common BFS estimation schemes for BFS extraction from single-peak Brillouin gain spectrum are simulated and analyzed,including quadratic fitting,Lorentz fitting and Lorentz crosscorrelation,and their limiting conditions and optimization schemes are analyzed.The fitting window size is very important for BFS extraction scheme based on fitting.For cross-correlation,The possibility of applying CRP in BFS extraction was analyzed.The correlation between reference spectrum and Brillouin gain spectrum was calculated by subsection,the CRP matrix was determined and the recursive graph was drawn,and the final BFS was obtained by centroid analysis,which could be regarded as subsection cross-correlation in principle.Finally,by analyzing the performance of each algorithm at different SNR and sampling frequency,and averaging the results for 1000 times,it is concluded that CRP achieves better BFS uncertainty under the adverse conditions of low SNR and large sampling frequency.When SNR=3dB,the uncertainty of 3.61MHz is obtained,and when SNR=5MHz,the uncertainty of 6.22MHz is obtained.(2)The Brillouin scattering characteristics of anti-parabolic graded refractive index low-mode fiber are analyzed and simulated.Discusses the traditional unimodal BFS extraction scheme in multimodal possibility of applying the Brillouin gain spectrum,including BFS extraction scheme based on fitting can be solved by piecewise Lorentz fitting.The BFS extraction scheme based on correlation can still be used in multi-peak scattering through mathematical model analysis,but it is no longer suitable for multi-peak Brillouin scattering spectrum generated by some optical fibers due to the spectrum broadening of Lorentz cross-correlation(the sum of different peak center frequencies is less than the sum of two times of bandwidth).CRP can still capture the central frequency information of different peaks in multi-peak Brillouin scattering spectrum due to its multifeature tracking characteristic.Finally,the performance of BP neural network in multi-peak data acquisition is analyzed.The ideal multi-peak Brillouin gain spectrum under different frequency shifts is simulated as training data,and then 100 groups of test data are selected to test the final system performance.The results show that the performance of BP neural network is better than Lorentz fitting.