Research On OTDR Signal Processing Algorithm

In recent years,with the rapid development of optical communication,the scale of optical network is also growing.If abnormalities such as breakage and damage occur somewhere in the optical line,the service of the optical network will be seriously affected.In order to ensure the quality of optical network communication service,anomaly locations need to be determined quickly in long-distance optical fibers so that anomalies can be resolved in time.This makes it very difficult to maintain the optical network.OTDR is a common tool in traditional optical fiber fault detection,but it is necessary to judge manually from the OTDR signal curve whether the test optical fiber has a fault and the type of failure.Due to the inefficiency and low accuracy of manual judgment,and the high requirement for experience of maintenance personnel,traditional optical maintenance methods can not meet the maintenance requirements of existing optical network gradually.In order to solve the limitation of artificial judgment fault,it is necessary to study OTDR signal processing algorithm,which can replace manual judgment.In this paper,OTDR signal processing algorithm is studied,the main research contents are as follows.(1)Wavelet transform threshold denoising based on grasshopper optimization algorithm.Combined with the characteristics of the return signal curve of OTDR test,the Time-Frequency Peak filter,the modulus-maximum denoising method based on the wavelet transform and the threshold denoising method based on the wavelet transform are used to denoise the OTDR signal.The denoising performance of the three denoising algorithms is compared from the root mean square error of the denoised signal and the offset of the incident location in the denoised signal.It is found that the threshold denoising method based on wavelet transform is more suitable for OTDR signal denoising.Since the threshold denoising method based on wavelet transformation needs to determine the wavelet basis function,decomposition level,threshold selection criteria,threshold usage method and threshold scaling method,this paper puts forward a threshold denoising method based on grasshopper optimization algorithm.The optimal parameters are selected to denoise the OTDR signal.The SNR of the denoised OTDR signal is raised from 16.92 d B to 34.47 d B compared with the noisy OTDR signal.(2)Optimal wavelet reconstruction reflection event detection method based on grasshopper optimization algorithm.When using wavelet analysis to detect reflected events,the wavelet basis function used in signal decomposition and the selected high frequency coefficient of reconstructed signals are the key factors that affect the accuracy of reflected events detection.In this paper,an optimal wavelet reconstructed reflective event detection method based on locust optimization is presented.The optimal wavelet basis function is used to decompose the denoised OTDR signal,the optimal selection of the detail coefficient of the wavelet transform is used for signal reconstruction,the optimal threshold is used to threshold the reconstructed signal,and the reconstructed signal with threshold processing is used to detect reflective events accurately.Reflective event detection using optimal wavelet reconstruction based on grasshopper optimization algorithm improves the recognition accuracy of reflective events by an average of 1.52% compared to using wavelet basis function selected directly in denoising.Non-reflective events are detected using a combination of two-point and least squares methods.A two-point method is used to locate the location where an event may occur.Data points around the location are fitted by least squares method,and the slope changes before and after the location are used to determine whether an event has occurred.Event recognition accuracy averaged 67.28%.(3)Increase the accuracy of event detection based on convolution neural network.For the detected reflective and non-reflective event intervals,the convolution neural network is used to identify them again to remove the misjudged events in the results.Further recognition by convolution neural network improved the recognition accuracy of reflective events by 6.48% and non-reflective events by 18.95%.