Research On Methods For Improving SNR Of RBOTDA System

Due to the advantages of single-ended operation,the Brillouin Optical TimeDomain Analysis(RBOTDA)system based on Rayleigh scattering has a wide range of applications in large-scale monitoring occasions.But the technology is also defective.The low signal-to-noise ratio of the system limits the measurement accuracy and sensing distance because of the weak Rayleigh scattering light.How to improve system SNR is one of the key problems of RBOTDA system.In this paper,a variety of noise reduction algorithms are applied to the RBOTDA system,and a two-dimensional lifting wavelet noise reduction algorithm is proposed.The RBOTDA experimental system is built in this paper.The traditional experimental data processing method is the cumulative average algorithm.In this paper,the cumulative average algorithm is analyzed based on the Matlab simulation platform.It is found that the signal-to-noise ratio of the measured signal increased with the number of accumulations.The signal-to-noise ratio reached 13.5549 dB at 10 000 accumulative averages,which is 7 dB higher than 256 times,at the expense of 9744 more accumulative averages.Accumulative averaging algorithm improves signal-tonoise ratio at the expense of real-time performance.The measurement time is long and real-time performance is poor.Wavelet transform is commonly used in signal noise reduction because of good time-frequency characteristics.Common wavelet noise reduction algorithms include modulus maximum noise reduction algorithm,spatial correlation filtering algorithm and wavelet threshold noise reduction algorithm.In this paper,one-dimensional Matlab simulation of RBOTDA experimental data was carried out base on three methods.The results show that the wavelet threshold noise reduction algorithm has the best noise reduction effect.The one-dimensional wavelet noise reduction algorithm only removes the correlation of the experimental data in the time domain and neglects the correlation between different scanning frequencies.In this paper,the experimental data are arranged in a two-dimensional matrix according to the scanning frequency.The row vectors represent the data at different positions of the optical fiber at the same scanning frequency,which have spatial correlation.The column vector is the data at the same position of the fiber under different scanning frequencies,which has time correlation.The redundancy of experimental data in the time domain and the spatial domain can be removed at the same time with two-dimensional wavelet noise reduction algorithm.So a higher signal-to-noise ratio than the one-dimensional wavelet algorithm can be obtained.The wavelet transform used above is based on Fourier transform.The wavelet transform based on lifting scheme inherits the multi-resolution features of the traditional wavelet,gets rid of the Fourier transform and avoids the redundant computation based on convolution algorithm in the traditional wavelet,reducing the computational complexity.In this paper,two-dimensional lifting wavelet threshold algorithm is proposed,combing high signal-to-noise ratio of two-dimensional wavelet and low computational complexity of wavelet,and the parameters are optimized.In this paper,the cumulative average algorithm,one-dimensional wavelet threshold noise reduction algorithm and two-dimensional lifting wavelet threshold noise reduction algorithm are used for RBOTDA system experimental data.It indicated that the accumulative average algorithm achieves a signal-to-noise ratio of 13.5549 dB and mean square error of 0.3275 at 10000 accumulative averages.Real-time performance of accumulative average algorithm is poor.One-dimensional wavelet threshold method and two-dimensional lifting wavelet threshold method are based on experimental data of 256 cycles collected,whose real-time performances are much better than the cumulative average algorithm.The signal to noise ratio of system achieves 20.6583 dB and mean square error is 0.2473 by the one-dimensional wavelet threshold noise reduction algorithm.The signal to noise ratio increased to 25.5076 dB and the mean square error is 0.1688 by two-dimensional lifting wavelet.At the same time,the computational complexity of two-dimensional lifting decomposition is onethird less than that of one-dimensional wavelet decomposition.