CN105115622A - Denoising algorithm of fiber Raman temperature sensing system based on independent component analysis - Google Patents

Abstract

The present invention discloses a denoising algorithm of a fiber Raman temperature sensing system based on independent component analysis. For the problem of low signal to noise ratio in the fiber Raman temperature sensing system, rapid independent component analysis is used to carry out denoising processing on a Raman scattering signal, at the same time for the uncertainty in the independent component analysis process, a sinusoidal signal with a certain amplitude and phase is added at the front end of the Raman scattering signal, according to the amplitude phase change before and after sinusoidal signal processing, the amplitude phase of the Raman scattering signal is corrected, and thus the Raman scattering signal is completely restored.

Description

Based on the denoise algorithm of the fiber Raman temperature-sensing system of independent component analysis

Technical field

The present invention relates to Fibre Optical Sensor field, specifically a kind of denoise algorithm of the fiber Raman temperature-sensing system based on independent component analysis.

Background technology

Fiber Raman temperature-sensing system is a kind of novel sensor-based system based on the temperature sensitive effect of spontaneous Raman scattering and optical time domain reflection principle, has corrosion-resistant, the feature such as volume is little, electromagnetism interference, is widely used in the fields such as traffic, building and oil gas.But, the very faint and impact of the various noise of the system that is subject to due to spontaneous Raman scattering signal, useful signal is submerged among noise completely, the performance parameter such as real-time, measuring accuracy of system is difficult to be guaranteed, therefore, suitable denoise algorithm is selected extremely to be necessary to the signal to noise ratio (S/N ratio) improving system.Noise main manifestations in fiber Raman temperature-sensing system is white noise, the denoise algorithm applied in current system mainly contains superposed average denoise algorithm, Noise Elimination from Wavelet Transform algorithm, pulse code denoise algorithm and empirical mode decomposition denoise algorithm, but these algorithms all have the limitation of oneself, seeking new denoise algorithm to reach more preferably effect is also trend of the times.Independent component analysis is a kind of blind source signal separation method grown up in recent years, utilizes its redundancy cancellation characteristic effectively can realize the separation of signal noise, in fiber Raman temperature-sensing system, has good application prospect.

Summary of the invention

The object of this invention is to provide a kind of denoise algorithm of the fiber Raman temperature-sensing system based on independent component analysis, with solve prior art due to Raman scattering signal the faint thus problem be submerged in noise.

In order to achieve the above object, the technical solution adopted in the present invention is:

Based on the denoise algorithm of the fiber Raman temperature-sensing system of independent component analysis, it is characterized in that: utilize fast independent component analysis to carry out denoising to Raman scattering signal, simultaneously for the uncertainty in fast independent component analysis process, the sinusoidal signal of certain amplitude and phase place is added in Raman scattering signal front end, the amplitude phase place of Raman scattering signal is corrected according to the amplitude phase place change before and after sinusoidal signal process, thus recover Raman scattering signal completely, comprise the following steps:

(1), add in noisy Raman scattering signal front end and the sinusoidal signal of certain amplitude and a phase bit be spliced to form observation signal;

(2), to splicing by noisy Raman scattering signal and sinusoidal signal the observation signal obtained go average and do whitening processing, obtaining whitening matrix X=(x ₁, x ₂..., x _n), wherein X is the observation signal after average and whitening processing, x ₁, x ₂..., x _nfor the n be made up of n independent signal source linear hybrid ties up random observation signal;

(3), based on negentropy maximization theory, objective function F (w)=E [G (w is determined ^tx)]+c (|| w|| ²-1), wherein c represents constant, and w represents a line of piece-rate system matrix W, and G () is non-quadratic function, and choosing of its form is relevant with the kind of source signal;

(4), application Newton iteration method ask this objective function optimum solution, make n=0, random selecting initial vector w (0), wherein || w (0) ||=1;

(5), w, w (n+1)=E{Xg (w is adjusted ^t(n) X) }-E{g'(w ^t(n) X) } w (n), n=n+1, wherein g and g ' is respectively the derivative of G and g;

(6), normalization, w (n+1)=w (n+1)/|| w (n+1) ||;

(7) if algorithm is not restrained, then go to step (5); If algorithm convergence, then estimate an independent component, y=wX;

(8) the sinusoidal signal phase-amplitude of noisy Raman scattering signal front end, after process there occurs change, and the Raman scattering signal of the signal rear end simultaneously after FastICA denoising also needs to make corresponding process.

The denoising simulation result of the present invention to 20km fiber Raman temperature-sensing system shows, system signal noise ratio obtains the improvement of 6.09dB, achieves ideal denoising effect.

Accompanying drawing explanation

Fig. 1 is independent component analysis theory diagram.

Fig. 2 is FastICA algorithm flow chart.

Fig. 3 is FastICA algorithm denoising effect figure.

Fig. 4 is the FastICA algorithm denoising effect figure that namely the present invention adds front end correction.

Embodiment

As depicted in figs. 1 and 2, fast independent component analysis is utilized to carry out denoising to Raman scattering signal, simultaneously for the uncertainty in independent component analysis process, the sinusoidal signal of certain amplitude and phase place is added in Raman scattering signal front end, correct the amplitude phase place of Raman scattering signal according to the amplitude phase place change before and after sinusoidal signal process, thus recover Raman scattering signal completely.Comprise the following steps:

(1), add in noisy Raman scattering signal front end and the sinusoidal signal of certain amplitude and a phase bit be spliced to form observation signal;

(2), to splicing by noisy Raman scattering signal and sinusoidal signal the observation signal obtained go average and do whitening processing, obtaining whitening matrix X=(x ₁, x ₂..., x _n), wherein X is the observation signal after average and whitening processing, x ₁, x ₂..., x _nfor the n be made up of n independent signal source linear hybrid ties up random observation signal;

(3), based on negentropy maximization theory, objective function F (w)=E [G (w is determined ^tx)]+c (|| w|| ²-1), wherein c represents constant, and w represents a line of piece-rate system matrix W, and G () is non-quadratic function, and choosing of its form is relevant with the kind of source signal;

(4), application Newton iteration method ask this objective function optimum solution, make n=0, random selecting initial vector w (0), wherein || w (0) ||=1;

(5), w, w (n+1)=E{Xg (w is adjusted ^t(n) X) }-E{g'(w ^t(n) X) } w (n), n=n+1, wherein g and g ' is respectively the derivative of G and g;

(6), normalization, w (n+1)=w (n+1)/|| w (n+1) ||;

(7) if algorithm is not restrained, then go to step (5); If algorithm convergence, then estimate an independent component, y=wX;

(8) the sinusoidal signal phase-amplitude of noisy Raman scattering signal front end, after process there occurs change, and the Raman scattering signal of the signal rear end simultaneously after FastICA denoising also needs to make corresponding process.

Simulation result and analysis:

The mathematical model of fiber Raman temperature-sensing system is such as formula described, and wherein S is backscattering coefficient, α _sfor the back scattering factor, v is light group velocity in a fiber, P ₀for incident optical power, T is light impulse length, and α is Transmission loss.The computing machine configuring Intel (R) Core (TM) i3-2130CPU3.40GHz processor and 6G internal memory utilizes Matlab to carry out emulation experiment, if systematic survey length is 20km, peak power is 100mW, and pulse width is 10ns.

$P\left(t\right)=\frac{1}{2}v\·S\·{\mathrm{\α}}_S\·P_0\·T\·e^{-\mathrm{\α}\·v\·t}---\left(8\right)$

When using FastICA algorithm denoising, the condition that measuring-signal number is more than or equal to source signal number need be met, when the noisy Raman scattering signal of process one dimension, virtual observation signal must be introduced one-dimensional signal is extended to multi-dimensional signal, the virtual observation signal herein chosen is added white Gaussian noise, and it and noisy Raman scattering signal form a two-dimensional observation signal jointly.Utilize FastICA algorithm to carry out denoising to the noisy Raman scattering signal of one dimension, result as shown in Figure 3.Can find out: 1) denoising effect of FastICA algorithm is better, and Raman scattering signal and part white Gaussian noise are successfully separated; 2) due to the uncertainty of independent component analysis (ICA), the amplitude of the Raman scattering signal after being separated and phase place is caused to change all to some extent.

For the deficiency of FastICA algorithm, introduce and add the FastICA algorithm that front end corrects, as previously mentioned, the condition that measuring-signal number is more than or equal to source signal number need be met, and the signal detected by Photoelectric Detection module mainly comprises Raman scattering signal and white noise, therefore can think that independent source here has 2.In addition, the interval time of fiber Raman temperature-sensing system Emission Lasers pulse is very short, and change between adjacent signals is very little, so get 4 groups of noisy Raman scattering signal here continuously to carry out ICA analysis as measuring-signal.Utilize the FastICA algorithm adding front end correction to carry out denoising to the noisy Raman scattering signal of the four-dimension, result as shown in Figure 4.Can find out: the denoising effect 1) adding the FastICA algorithm that front end corrects is better, and Raman scattering signal and part white Gaussian noise are successfully separated; 2) by the change of amplitude and phase place before and after the sinusoidal signal process of signal front end, Raman scattering signal can be recovered completely; 3) change through the temperature field except 10km place corresponding on the Raman scattering signal after making an uproar is high-visible.

In order to evaluate the denoising effect adding the FastICA algorithm that front end corrects, calculate the signal to noise ratio (S/N ratio) before and after noisy Raman scattering signal denoising respectively, as shown in table 1.

Table 1FastICA innovatory algorithm is to the improvement of system signal noise ratio

Claims (1)

1. based on the denoise algorithm of the fiber Raman temperature-sensing system of independent component analysis, it is characterized in that: utilize fast independent component analysis FastICA to carry out denoising to Raman scattering signal, simultaneously for the uncertainty in fast independent component analysis process, the sinusoidal signal of certain amplitude and phase place is added in Raman scattering signal front end, the amplitude phase place of Raman scattering signal is corrected according to the amplitude phase place change before and after sinusoidal signal process, thus recover Raman scattering signal completely, comprise the following steps:

(1), add in noisy Raman scattering signal front end and the sinusoidal signal of certain amplitude and a phase bit be spliced to form observation signal;

(2), to splicing by noisy Raman scattering signal and sinusoidal signal the observation signal obtained go average and do whitening processing, obtaining whitening matrix X=(x ₁, x ₂..., x _n), wherein X is the observation signal after average and whitening processing, x ₁, x ₂..., x _nfor the n be made up of n independent signal source linear hybrid ties up random observation signal;

(3), based on negentropy maximization theory, objective function F (w)=E [G (w is determined ^tx)]+c (|| w|| ²-1), wherein c represents constant, and w represents a line of piece-rate system matrix W, and G () is non-quadratic function, and choosing of its form is relevant with the kind of source signal;

(4), application Newton iteration method ask this objective function optimum solution, make n=0, random selecting initial vector w (0), wherein || w (0) ||=1;

(5), w, w (n+1)=E{Xg (w is adjusted ^t(n) X) }-E{g'(w ^t(n) X) } w (n), n=n+1, wherein g and g ' is respectively the derivative of G and g;

(6), normalization, w (n+1)=w (n+1)/|| w (n+1) ||;

(7) if algorithm is not restrained, then go to step (5); If algorithm convergence, then estimate an independent component, y=wX;

(8) the sinusoidal signal phase-amplitude of noisy Raman scattering signal front end, after process there occurs change, and the Raman scattering signal of the signal rear end simultaneously after FastICA denoising also needs to make corresponding process.