CN106788697A - A kind of noise-reduction method of phase sensitive OTDR signals - Google Patents

Abstract

The invention relates to a noise reduction method for a phase-sensitive OTDR signal. The invention collects and superimposes the phase-sensitive OTDR sensing curve multiple times, takes the composed two-dimensional matrix as the processing object, performs positive circular translation on it, and uses fast discrete curvelet transform to perform multi-scale conversion on the signal after the positive circular translation. Decompose, analyze and reconstruct the components of each scale after threshold value processing, so as to suppress the background noise and achieve the effect of noise reduction, so as to observe the position of the real disturbance. The noise reduction method of the phase-sensitive OTDR signal in the present invention determines the threshold value according to the sensing curve itself, and uses different threshold values for each scale layer, which can well attenuate random noise and realize the separation of signal and noise to the greatest extent.

Description

A Noise Reduction Method for Phase Sensitive OTDR Signals

technical field

The invention relates to a noise reduction method for a phase-sensitive OTDR signal, and belongs to the technical field of optical fiber sensing and detection.

Background technique

The safety of national defense, military, civilian facilities and people's lives and properties is a major issue related to the national economy and people's livelihood. Therefore, my country has major technical needs in the fields of perimeter security, long-distance pipeline safety, and large-scale structural health monitoring. Phase-sensitive OTDR (Optical Time Domain Reflectometer) is an emerging representative of distributed optical fiber sensing technology. It has high sensitivity, light weight, small size, no electromagnetic interference, and can continuously detect the spatial distribution of parameters such as strain and vibration on the transmission path. and time-varying information. In recent years, it has been widely used in vibration measurement in petroleum, transportation, structure and other fields.

The phase-sensitive OTDR mainly detects the interference effect of Rayleigh scattered light, and it is necessary to avoid rapid changes in optical power. In fact, a slight disturbance in the outside world will cause a change in the optical phase, which will cause a drastic change in the detected optical power, causing the real signal to be submerged in the noise. At the same time, random changes in detection results caused by polarization fading in optical fibers may also be recognized by OTDR as real signals. Therefore, identifying real signals, reducing background noise, and improving detection performance are problems that need to be solved urgently in phase-sensitive OTDR applications.

Chinese patent CN102946271A discloses a method and device for noise reduction of OTDR test curves. After the point sequence of the OTDR test curve is subjected to discrete Fourier transform in time domain transform and frequency domain through the time domain transform frequency domain module, in the frequency domain The OTDR curve is processed by low-pass filtering through the low-pass filtering module, and the high-frequency part in the curve is filtered to obtain the curve of the low-frequency part, and then the filtered curve is subjected to discrete Fourier inverse through the frequency domain inversion time domain module Transform, restore the curve to the OTDR curve in the time domain, so that the noise in the OTDR test curve can be reduced. The processing object of the method and device disclosed in this patent document is a sensing curve, and there are chances in the actual process, which can easily lead to false positives or false negatives. Secondly, the processing of the sensing curve by the method and device is relatively simple. In the case of high frequency and low frequency, it is easy to lose the effective part of the signal, resulting in a decrease in the signal-to-noise ratio, or even no signal detection.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a noise reduction method for phase-sensitive OTDR signals.

Terminology Explanation:

Backward Rayleigh scattering: In the optical fiber, due to the random fluctuation of the optical fiber density, the local fluctuation of the refractive index causes light to scatter in all directions. The wavelength of the scattered light of Rayleigh scattering is equal to the wavelength of the incident light, and there is no frequency change. Backward Rayleigh scattering refers to scattered light directed towards the incident end.

Fast Discrete Curvelet Transform: It belongs to the second-generation Curvelet Transform theory, which divides the frequency domain of the processing object by concentric square areas, and processes each sub-block separately.

Summary of the invention:

The present invention adopts the following technical scheme: the phase-sensitive OTDR sensing curve is collected and superimposed multiple times, and the two-dimensional matrix formed is used as the processing object, and its positive cycle is translated, and the positive cycle is translated by fast discrete curvelet transform. The multi-scale decomposition of the signal is carried out, and the components of each scale are analyzed and reconstructed after threshold value processing, so as to suppress the background noise and achieve the effect of noise reduction, so as to observe the position of the real disturbance.

Technical scheme of the present invention is:

A noise reduction method for a phase-sensitive OTDR signal is used for the denoising of a Rayleigh scattered light signal, comprising the following steps:

1) Repeat the process of "positive cyclic translation - fast discrete curvelet transform - threshold processing - fast discrete curvelet inverse transform - reverse cyclic translation", the specific formula is as follows:

in, is the sensing curve matrix after noise reduction processing, S is the sensing curve matrix, F _{n, n} are positive circular translation operators, F _{-n, -n} are reverse circular translation operators, I and I ^-1 are fast Discrete curvelet transform operator and fast discrete curvelet inverse transform operator, T is the threshold reconstruction operator, n ₁ and n ₂ are the translation of the sensing curve matrix in the direction of row and column, N ₁ is the sensor The translation range of the curve matrix in the row direction, N ₂ is the translation range of the sensing curve matrix in the column direction; the sensing curve matrix is translated in the direction of the row and column, each time a row is moved, and the maximum number of translations is the number of rows multiplied by With the number of columns, the shifted signal is denoised and shifted in reverse, and the average is taken after multiple cycles of shifting, which effectively eliminates the ringing effect.

The translation of the sensing curve matrix in the row and column directions corresponds to the translation of the superimposed sensing curves in the horizontal and vertical directions; since the curvelet transform does not have translation invariance, the adjacent positions of the discontinuous points of the effective signal will produce false Gibbs phenomenon, while the positive cyclic translation effectively suppresses the pseudo-Gibbs phenomenon.

Preferably according to the present invention, the noise reduction method of the phase-sensitive OTDR signal comprises specific steps as follows:

1.1) The phase-sensitive OTDR signal curve is collected and superimposed for 100 to 1000 times (in the example, the result of collecting 1000 times of superimposition), and the superimposed sensing curve is expressed as an M×N sensing curve matrix;

1.2) positive circular translation of the sensing curve matrix; translation of the sensing curve matrix in the direction of rows and columns, one row and one column each time, the maximum number of times of translation is the number of rows multiplied by the number of columns; positive circular translation is a technology in the art The means of data processing familiar to the personnel.

1.3) Scale-decompose the sensory curve matrix after positive circular translation by fast discrete curvelet transform, and obtain the curvelet coefficient matrix C(j,l,k); where j is the scale parameter for decomposition, and l is each scale The direction parameter corresponding to the parameter, k is the position parameter corresponding to each direction parameter;

1.4) Perform threshold processing on the curvelet transform coefficients corresponding to different scale layers to suppress background noise:

Among them, C ^r (j,l,k) is the thresholded curvelet coefficient matrix, and T is the threshold coefficient; for the coefficient matrix C(j,l,k), different scale layers represent different frequency components, and for each The scale layer is processed with different threshold coefficients to preserve the signal component and remove the noise component.

Curvelet transform can be divided into three scale layers: COARSE layer, DETAIL layer and FINE layer. The DETAIL layer can be further subdivided into many layers. Generally, the default ceil(log2(min(M,N))-3) Divide so many layers (M, N is the matrix size). The innermost layer, the coarse layer, mainly contains some low-frequency information and does not have directionality. The outermost fine layer is mainly some high-frequency information. Each layer of it contains many directions, and the more layers there are, the finer the directions are divided.

1.5) performing fast discrete curvelet inverse transform and inverse cyclic translation on the thresholded curvelet coefficient matrix; wherein, fast discrete curvelet inverse transform and inverse cyclic translation are the inverse processes of fast discrete curvelet transform and positive cyclic translation respectively.

1.6) Repeat steps 1.2)-1.5) (4 times in the example) to obtain the sensory curve matrix after noise reduction processing. The number of times to repeat steps 1.2)-1.4) is determined according to the range of cyclic translation.

Further preferably, in the step 1.3), the fast discrete curvelet transform is used to decompose the scale, and the specific method for obtaining the curvelet coefficient matrix C (j, l, k) is: utilize the existing curvelet toolbox of matlab to sense Perform fast discrete curvelet transform on the curve matrix to obtain the curvelet coefficient matrix C(j,l,k); the implementation method of fast discrete curvelet inverse transform is to use the existing curvelet toolbox of matlab to realize. Using matlab to perform fast discrete curvelet transform and fast discrete curvelet inverse transform is a well-known technical means in the prior art.

Further preferably, in the step 1.3), the number of layers of scale decomposition:

[J]=log ₂ (M,N)-3

Among them, [J] represents the integer part of J; the first layer is the Coarse scale layer, which is a matrix composed of low-frequency coefficients; the outermost layer is the Fine scale layer, which is a matrix composed of high-frequency coefficients; the middle layer is the Detail scale layer , is a matrix composed of medium and high frequency coefficients.

Further preferably, in the step 1.4), the threshold coefficient is calculated using the Monte Carlo threshold method:

T＝k e _j e

Among them, e _j is the coefficient standard deviation obtained by Monte Carlo test after fast discrete curvelet transform of Gaussian white noise with mean value 0 and variance 1; e is the noise standard deviation in the phase-sensitive OTDR signal;

j=1,2,3...[J]; L _j is the number of directions in the scale layer; k is a scale-dependent coefficient, and for different scales, k takes different values to meet the needs. The specific process of the coefficient standard deviation obtained by the Monte Carlo test is to obtain the norm of the coefficient matrix in each scale and direction, and then divide it by the number of elements contained in the matrix.

Preferably according to the present invention, said step 1) also includes the step of performing differential processing on the noise reduction result:

Among them, x(n) represents the phase-sensitive OTDR curve after noise reduction, y(n) represents the phase-sensitive OTDR curve after differential processing, and N represents the total number of phase-sensitive OTDR curves. Differential processing can more intuitively see the location of the disturbance.

Further preferably, in the step 1.1), the number of times of collecting and superimposing the phase-sensitive OTDR signal curve is 100 to 1000 times.

The beneficial effects of the present invention are:

1. The noise reduction method of the phase-sensitive OTDR signal of the present invention determines the threshold value according to the sensing curve itself, and uses different threshold values for each scale layer, which can well attenuate random noise; realize signal and noise to the greatest extent separation;

2. The noise reduction method of the phase-sensitive OTDR signal of the present invention does not require the frequency of the processed signal, effectively avoids the loss of the effective part of the signal, enhances the signal-to-noise ratio, and improves the accuracy of external disturbance vibration source positioning;

3. The noise reduction method of the phase-sensitive OTDR signal of the present invention, the phase-sensitive OTDR curve is collected multiple times and superimposed and then denoised, which greatly reduces the contingency in the actual process and effectively avoids false alarms or omissions;

4. The noise reduction method of the phase-sensitive OTDR signal of the present invention effectively improves the accuracy of detecting disturbance based on the phase-sensitive OTDR system, improves the detection performance of working in a complex noise environment, and can be widely applied to pipeline transportation, bridge detection, etc. field.

Description of drawings

Fig. 1 is the flow chart of the denoising method of phase-sensitive OTDR signal of the present invention;

Fig. 2 is the schematic diagram of the phase-sensitive OTDR device described in embodiment 2;

Fig. 3 is the overlay diagram of the signal curve of piezoelectric ceramics simulated vibration;

Fig. 4 is the signal curve overlay diagram after curvelet transform noise reduction;

FIG. 5 is an overlay diagram of signal curves after differential processing.

FIG. 6 is a superposition diagram of signal curves directly subjected to differential processing without noise reduction processing.

detailed description

The present invention will be further described below in conjunction with the embodiments and the accompanying drawings, but is not limited thereto.

Example 1

As shown in Figure 1.

A noise reduction method for a phase-sensitive OTDR signal is used for the denoising of a Rayleigh scattered light signal, comprising the following steps:

1) Repeat the process of "positive cyclic translation - fast discrete curvelet transform - threshold processing - fast discrete curvelet inverse transform - reverse cyclic translation", the specific formula is as follows:

in, is the sensing curve matrix after noise reduction processing, S is the sensing curve matrix, F _{n, n} are positive circular translation operators, F _{-n, -n} are reverse circular translation operators, I and I ^-1 are fast Discrete curvelet transform operator and fast discrete curvelet inverse transform operator, T is the threshold reconstruction operator, n ₁ and n ₂ are the translation amounts of the sensing curve matrix in the row and column direction respectively, N ₁ is the sensor The translation range of the curve matrix in the row direction, N ₂ is the translation range of the sensing curve matrix in the column direction; the sensing curve matrix is translated in the direction of the row and column, each time moving one row and one column, and the maximum number of translations is the number of rows multiplied by With the number of columns, the shifted signal is denoised and shifted in reverse, and the average is taken after multiple cycles of shifting, which effectively eliminates the ringing effect.

The translation of the sensing curve matrix in the row and column directions corresponds to the translation of the superimposed sensing curves in the horizontal and vertical directions; since the curvelet transform does not have translation invariance, the adjacent positions of the discontinuous points of the effective signal will produce false Gibbs phenomenon, while the positive cyclic translation effectively suppresses the pseudo-Gibbs phenomenon.

Example 2

The noise reduction method of the phase-sensitive OTDR signal as described in embodiment 1, the difference is that the noise reduction method of the phase-sensitive OTDR signal comprises specific steps as follows:

1.1) The phase-sensitive OTDR signal curve is collected and superimposed 1000 times, and the superimposed sensing curve is expressed as a sensing curve matrix of 1000×1000;

1.2) Perform positive circular translation on the sensing curve matrix; translate the sensing curve matrix in the direction of rows and columns, one row and one column at a time, and the maximum number of translations is the number of rows multiplied by the number of columns.

1.3) Scale-decompose the sensory curve matrix after positive circular translation by fast discrete curvelet transform, and obtain the curvelet coefficient matrix C(j,l,k); where j is the scale parameter for decomposition, and l is each scale The direction parameter corresponding to the parameter, k is the position parameter corresponding to each direction parameter;

1.4) Perform threshold processing on the curvelet transform coefficients corresponding to different scale layers to suppress background noise:

Among them, C ^r (j,l,k) is the curvelet coefficient matrix after thresholding, and T is the threshold coefficient; for the coefficient matrix C(j,l,k), different scale layers represent different frequency components, and for each The scale layer is processed with different threshold coefficients to preserve the signal component and remove the noise component.

1.5) performing fast discrete curvelet inverse transform and inverse cyclic translation on the thresholded curvelet coefficient matrix; wherein, fast discrete curvelet inverse transform and inverse cyclic translation are the inverse processes of fast discrete curvelet transform and positive cyclic translation respectively.

1.6) Repeat steps 1.2)-1.5) 4 times to obtain the sensing curve matrix after noise reduction processing.

In this embodiment, the schematic diagram of the phase-sensitive OTDR device used is shown in Figure 2. The number of acquisitions is 1000, the length of the optical fiber is 1Km, and the analog vibration signal is a 200HZ sinusoidal signal, which is set at a position of 880m. The overlay diagram of the signal curve of the simulated vibration of the piezoelectric ceramic is shown in Fig. 3 . The sensing signal curve matrix is a 1000×1000 matrix, and the positive circular translation is performed on the matrix, and the number of cycles is 4, that is, two rows and two columns are translated to obtain the positive circular translation matrix.

The overlay of the signal curve after denoising by the curvelet transform is shown in Figure 4.

Example 3

The noise reduction method of the phase-sensitive OTDR signal as described in embodiment 2, the difference is that in the step 1.3), the fast discrete curvelet transform is used to perform scale decomposition to obtain the curvelet coefficient matrix C (j, l, k The specific method of ) is: use the existing curvelet toolbox of matlab to perform fast discrete curvelet transform on the sensing curve matrix to obtain the curvelet coefficient matrix C(j,l,k); the realization method of fast discrete curvelet inverse transform is, It is realized by using the existing curvelet toolbox of matlab.

Fast discrete curvelet inverse transform is realized by wrapping algorithm:

A. For two-dimensional objective function Do two-dimensional Fourier transform to get its two-dimensional frequency domain representation:

B. In the frequency domain, for each pair of scales and angles (j,l), obtain and the product of

C. Wrap the data obtained in the previous step around the origin to get

D. Yes Perform a two-dimensional inverse Fourier transform to obtain the curvelet coefficient matrix C(j,l,k).

Example 4

The noise reduction method of the phase-sensitive OTDR signal as described in Embodiment 2, the difference is that in the step 1.3), the number of layers of scale decomposition:

[J]=log ₂ (M,N)-3

Among them, [J] represents the integer part of J; the first layer is the Coarse scale layer, which is a matrix composed of low-frequency coefficients; the outermost layer is the Fine scale layer, which is a matrix composed of high-frequency coefficients; the middle layer is the Detail scale layer , is a matrix composed of medium and high frequency coefficients.

In this embodiment, the sensing curve matrix is decomposed by 7 layers of fast discrete curvelet transform. For the j=1 scale layer, the direction parameter l=1, that is, no direction information; for the j=2 scale layer, the direction parameter l={1,2,...,16}, which includes 16 direction subbands; for j= For the 3-scale layer, the direction parameter l={1,2,...,32}, which includes 32 direction sub-bands; for the j=4-scale layer, the direction parameter l={1,2,...,32}, which includes 32 direction sub-bands; for j=5 scale layer, direction parameter l={1,2,...,64}, which includes 64 direction sub-bands; for j=6 scale layer, direction parameter l={1,2, ...,64}, that is, it contains 64 direction sub-bands; for j=7 scale layer, the direction parameter l={1,2,...,128}, that is, it contains 128 direction sub-bands.

Example 5

The noise reduction method of the phase-sensitive OTDR signal as described in Embodiment 4, the difference is that in the step 1.4), the threshold coefficient is calculated by the Monte Carlo threshold method:

T＝k e _j e

Among them, e _j is the coefficient standard deviation obtained by Monte Carlo test after fast discrete curvelet transform of Gaussian white noise with mean value 0 and variance 1; e is the noise standard deviation in the phase-sensitive OTDR signal;

j=1,2,3...[J]; L _j is the number of directions in the scale layer; k is a scale-dependent coefficient, and for different scales, k takes different values to meet the needs. The specific process of the coefficient standard deviation obtained by the Monte Carlo test is to obtain the norm of the coefficient matrix in each scale and direction, and then divide it by the number of elements contained in the matrix.

Example 6

The noise reduction method of the phase-sensitive OTDR signal as described in embodiment 1, the difference is that after the step 1) also includes

Including the steps of differential processing of the noise reduction results:

Among them, x(n) represents the phase-sensitive OTDR curve after noise reduction, y(n) represents the phase-sensitive OTDR curve after differential processing, and N represents the total number of phase-sensitive OTDR curves. Differential processing can more intuitively see the location of the disturbance.

In this embodiment, the total number of sensing curves is 1000, and the process of differential processing is to make a difference between the first sensing curve and the second sensing curve, and the result is used as the first column of the new matrix, and the second sensing curve and The difference between the third sensing curve is taken as the second column of the new matrix, and so on, the difference between the 1000th sensing curve and the first sensing curve is taken as the 1000th column of the new matrix, the result is shown in the figure 5.

Comparative example:

For the signal curve of the simulated vibration of piezoelectric ceramics, the differential processing is directly performed without noise reduction processing; the differential processing process is the same as the differential process in Embodiment 6; the overlay diagram of the obtained signal curve is shown in FIG. 6 .

From the comparison of Figure 5 and Figure 6, it can be seen that after the denoising results of curvelet transform threshold processing, it can be clearly seen that there is vibration information at 880m, and random noise is filtered out, and the signal-to-noise ratio is higher. Without denoising processing, the noise component is obvious, the signal is submerged in the noise, and no vibration information can be seen at all. Therefore, it is proved that the present invention has beneficial effects of reducing random noise, improving signal-to-noise ratio, and accurately locating disturbance in the process of phase-sensitive OTDR signal processing.

Claims (7)

1. A method for noise reduction of a phase sensitive OTDR signal, comprising the steps of:

1) the process of 'forward cyclic translation, fast discrete curvelet transformation, threshold processing, fast discrete curvelet inverse transformation and inverse cyclic translation' is repeated, and the specific formula is as follows:

$\hat{S}=\frac{1}{N_1{N}_2}\underset{n_1=0,n_2=0}{\overset{N_1,N_2}{\Σ}}{F}_{-n,-n}\left(I^{-1}\right(TI\left(F_{n,n}\left(S\right)\right)\left)\right)$

wherein,is a sensing curve matrix after noise reduction treatment, S is a sensing curve matrix, F_n,nFor positive cyclic translation operators, F_-n,-nFor the inverse cyclic translation operator, I and I^-1Respectively a fast dispersion curvelet transform operator and a fast dispersion curvelet inverse transform operator, T being a threshold reconstruction operator, n₁And n₂Respectively the amount of translation of the matrix of the sensing curve in the row and column directions, N₁For the range of translation of the matrix of the sensing curve in the row direction, N₂The translation range of the sensing curve matrix in the column direction is shown.

2. The method of claim 1, wherein the method of reducing noise in a phase-sensitive OTDR signal comprises the following steps:

1.1) carrying out multiple acquisition and superposition on a phase-sensitive OTDR signal curve, and representing the superposed sensing curve as an MXN sensing curve matrix;

1.2) carrying out positive cycle translation on the sensing curve matrix;

1.3) carrying out scale decomposition on the sensing curve matrix subjected to positive cycle translation by adopting rapid discrete curvelet transformation to obtain a curvelet coefficient matrix C (j, l, k); wherein j is a decomposed scale parameter, l is a direction parameter corresponding to each scale parameter, and k is a position parameter corresponding to each direction parameter;

1.4) carrying out threshold processing on curvelet transform coefficients corresponding to different scale layers, and inhibiting background noise:

$C^r(j,l,k)=\left\{\begin{array}{cc}C(j,l,k)& C(j,l,k)\&GreaterEqual;T\\ 0& C(j,l,k)<T\end{array}\right.$

wherein, C^r(j, l, k) is a curvelet coefficient matrix after threshold processing, and T is a threshold coefficient;

1.5) carrying out fast discrete curvelet inverse transformation and inverse cyclic translation on the curvelet coefficient matrix after threshold processing;

1.6) repeating the steps 1.2) -1.5) to obtain a sensing curve matrix after noise reduction treatment.

3. The method for reducing noise of a phase-sensitive OTDR signal according to claim 2, characterized in that, in said step 1.3), the specific method for obtaining the curvelet coefficient matrix C (j, l, k) by performing the scale decomposition using the fast discrete curvelet transform is: carrying out rapid discrete curvelet transformation on the sensing curve matrix by using the conventional curvelet tool box of matlab to obtain a curve coefficient matrix C (j, l, k); the method for realizing the rapid dispersion curvelet inverse transformation is realized by utilizing the existing curvelet tool box of matlab.

4. A method of noise reduction of a phase sensitive OTDR signal according to claim 2, characterized in that, in said step 1.3), the number of layers of the scale decomposition:

[J]＝log₂(M,N)-3

wherein [ J ] represents the integer part of J; the first layer is a Coarse scale layer and is a matrix consisting of low-frequency coefficients; the outermost layer is a Fine scale layer and is a matrix consisting of high-frequency coefficients; the middle layer is a Detail scale layer and is a matrix composed of medium and high frequency coefficients.

5. A method for noise reduction of phase sensitive OTDR signals according to claim 4, wherein in step 1.4), said threshold coefficients are calculated using a Monte Carlo threshold method:

T＝k·e_j·e

wherein e is_jAfter fast discrete curvelet transformation is carried out on Gaussian white noise with the mean value of 0 and the variance of 1, a Monte Carlo test is carried out to obtain a coefficient standard deviation; e is the noise standard deviation in the phase sensitive OTDR signal;

$k=2^{\frac{1-j}{2}}\sqrt{2{\log }_2\left(\frac{L_j}{\&lsqb;J\&rsqb;}\right)}$

j＝1,2,3…[J]；L_jis the number of directions of the scale layer.

6. A method of noise reduction of a phase sensitive OTDR signal according to claim 1, characterized in that, after said step 1), it further comprises the step of performing a difference processing on the noise reduction result:

$\begin{array}{c}y\left(n\right)=x\left(n\right)-x(n+1)\\ y\left(N\right)=x\left(N\right)-x\left(1\right)\end{array},(n=1,2,3...N-1)$

wherein, x (N) represents the phase-sensitive OTDR curve after noise reduction, y (N) represents the phase-sensitive OTDR curve after differential processing, and N represents the total number of the phase-sensitive OTDR curves.

7. The method for reducing noise of a phase-sensitive OTDR signal according to claim 2, characterized in that, the number of times of collecting and superimposing the phase-sensitive OTDR signal curve in step 1.1) is 100-1000.