SIGNAL LOSS DETECTION METHOD FOR DISTRIBUTED ACOUSTIC SENSING SYSTEMS

Abstract

A method that detects the location of undesired levels of signal loss for DAS systems. In the method, raw data is received from a DAS system with dual photodetector. Statistics are obtained from the received raw data, then processed according to the dual photodetector. The obtained statistical data are reconstructed to remove noise and the reconstructed signal is used to form the power statistics of the DAS signal in each channel. The power statistics are expected to be linearly decreasing with the distance from the sensor. A change detection algorithm is developed to detect the possible undesired levels of signal loss and to find the location of the signal loss.

Claims

1. A signal loss detection method for distributed acoustic sensing (DAS) systems, comprises steps of: collecting DAS data of channels from one time frame, down sampling the DAS data and collecting down sampled DAS data, calculating a standard deviation of each channel to get statistical power data from the down sampled DAS data, removing an optical coupler effect on the statistical power data, performing line fitting from each channel to two ends of the DAS system to find if an abrupt change in the power levels with respect to the channels occurs or not, selecting a best channel with minimum fit errors using a maximum likelihood estimation, calculating a difference between two line fits on the best channel, comparing the difference with a threshold, giving an alarm that there is a significant signal loss at the best channel if the threshold is exceeded.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0008] FIG. 1 shows block diagram of the present invention.

[0009] FIG. 2 shows received raw data from a DAS system with dual photodetector.

[0010] FIG. 3 shows the power statistics and the detection of power loss on channel 2150.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] Data Processing

[0012] Data pre-processing step comprises of the first five blocks in the FIG. 1. In the collect DAS data block, the DAS data of C channels from one time frame is collected. In the down sample collected DAS data block, N.sub.1 of the data obtained is down sampled. In the collect down sampled DAS data block, N.sub.2 vof the down sampled data is collected, meaning a total of (N.sub.1×N.sub.2) data is collected and used up to this point. In the create statistics of collected data block, the standard deviation of each channel is calculated to get a statistical power data from N.sub.2 data obtained. In the remove optical coupler effect on statistical data block, the effect of the coupler is negated by amplifying the merged signals at close-in distances. This is possible because of the Gaussian assumption. If a gaussian random variable is multiplied with a constant, the new random variable is also a Gaussian random variable with its standard deviation multiplied with the constant.

[0013] Signal Loss Detection

[0014] Signal loss detection step comprises of the remaining blocks. In the perform line fitting from each channel to both ends of DAS system block, the aim is to find if an abrupt change in the power levels with respect to channels occurs or not. The assumption here is, if a finite sample y.sub.1, . . . , y.sub.N assumed to have a probability density p(θ) and for each i, 1≤i≤N, the samples obtained reside in this probability distribution, there is no change in the distribution of the samples. However, for an index k, if there can be found θ.sub.0 and θ.sub.1, that θ=θ.sub.0 for 1≤i ≤k, θ=θ.sub.1 for k≤i ≤N, it can be said that there is a change in the distribution at position k. Let y.sub.k be a sequence of independent random observations with Gaussian distribution of (u.sub.k,σ) where u.sub.k=α+βk, 1≤k≤N. In this case θ=(α,β), θ.sub.0=(α.sub.0, β.sub.0), θ.sub.1=(α.sub.1, β.sub.1). This distribution explains the power statistics obtained by the last block with k being the channel and yk being the power of the k.sup.th channel. Therefore, at each channel k, maximum likelihood estimation can be used to find θ.sub.0 and θ.sub.1, which ultimately equals to fitting two lines for its lower channels (from 0 to k) and higher channels (from k to N), respectively.

[0015] In the select the best channel with minimum fit errors block, a maximum likelihood estimation is used to find the best channel that minimizes the errors when the two lines are fit to explain the data. Since the distribution is gaussian, it is equal to calculating sum of squared errors at each channel compared to lines.

[0016] In the calculate the difference between two line fits on the best channel block, the two parameters θ.sub.0 and θ.sub.1 used at the best channel c.sub.best and the difference between the lower channels fit and higher channels fit are calculated.

[0017] In the perform decision rule with threshold block, the calculated difference is compared with a threshold, and if the threshold is exceeded, the system gives an alarm that there is a significant signal loss at the specified channel c.sub.best.

[0018] Work Principle

[0019] The received raw data from the DAS system with dual photodetector can be seen in FIG. 2. Using this raw data, the data that would be obtained by the sensor without the dual photodetector can be reconstructed, but this will not be necessary for the algorithm hence it will not be calculated. Instead, statistics are obtained from the received raw data, then processed according to the dual photodetector.

[0020] Although the amplitude of the raw data gives information about the power level of the channels, it is highly affected from the interior and exterior noises. To get rid of those effects, the fluctuations in the signal at a specific channel with respect to time can be used. Then the obtained statistical data are reconstructed into a signal that would be obtained if the raw signal didn't put through a photodetector. The reconstructed statistical data is used to form the power statistics of the DAS signal in each channel. The power statistics are expected to be linearly decreasing with the distance from the sensor. A change detection algorithm is developed to detect the possible undesired levels of signal loss and to find the location of the signal loss.

[0021] It can be seen that there is a significant power loss in channel 2150 in FIG. 3. The algorithm chooses the hypothesis that there is a significant power loss in the system. The lines correspond to the optimum Gaussian distribution means with (α.sub.0+iβ.sub.0), (α.sub.1+iβ.sub.1), where i is the channel number.