Open Access
Add to BibSonomy
Abstract
A new type of dual Mach-Zehnder interferometer (DMZI) scheme is presented to eliminate the polarization induced fading usually encountered in fiber-optic intrusion detection systems that use DMZIs. With such a new optical scheme, two identical signal waveforms with only a time delay that corresponds to the location of intrusion can always be obtained without using a polarization control loop. In the study, we also present a new algorithm for locating the intrusion-induced disturbance on the fiber. The experiments carried out in our lab have demonstrated that locating errors of <26 m can be obtained with the presented system used for detecting an intrusion point on a 250 meter long fiber cable. Also, the proposed DMZI system has been tested for years without changing any optical components except the laser type and the length of the sensing fiber. Notably, the thresholds for determining the intrusion have never been altered since the laser type was changed in 2018. In addition, detections of the clockwise and counter-clockwise signals have maintained a condition of high interference visibility, and the locating capability has remained at the same level.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Fiber-optic intrusion detection (FID) systems have been widely used for detecting intruder-induced disturbance on the fiber cable set up on protected premises. Two kinds of FID techniques have been mostly employed for locating the disturbance on the sensing fiber. One kind is the optical time domain interferometer based technique, which has been widely used to detect and locate the disturbance-induced phase variation in Rayleigh backscattered light [1–5]. Capability of long-range detection and multipoint location has been demonstrated using this kind of technique; however this technique usually involves the use of relatively sophisticated circuits in event determination. Another kind of FID techniques use optical interferometers.
Various types of optical interferometers have been studied for the purpose of intrusion detection. A Sagnac loop interferometer was proposed more than three decades ago for disturbance detection along a 200 meter loop and subsequently applied for the same purpose [6–8]. Michelson interferometers were also used for distributed disturbance detection [9,10]. A system with merged Sagnac and Michelson interferometers [11,12], and that with combined Sagnac and Mach-Zehnder interferometers [13,14] were also developed for detecting and locating the disturbance applied on sensing fibers about two decades ago.
Meanwhile, dual Mach-Zehnder interferometers (DMZIs) have been used as disturbance sensors for intrusion detection. A traditional DMZI system uses a laser source with the light split into two paths for a clockwise and a counter-clockwise interferometers, respectively. A cross correlation algorithm is usually used for determining the time delay between the two time-domain signals detected in the two interferometers. Based on different positioning methods used for this type of interferometers, different research groups presented distinguishing capabilities of their sensor systems. An experiment was conducted to show a average measurement error of 390 meters for an 18.46 km long detection range [15]. Effect of polarization fading was eliminated through a polarization control to obtain a locating error of 160 meters for a 112 km fiber length [16]. To overcome the effect of polarization induced fading, an polarization control method based on chaotic particle swarm optimization algorithm was used and achieved a locating error of ±20 m over a 2.25 km sensing cable, which was one order of magnitude lower than that obtained by using the traditional polarization control method based on the criterion of interference visibility [17]. Another research group used wavelength-division multiplexing to reduce the influence of Rayleigh backscattering on the signals detected in the DMZIs. A locating error of 52.5 m was obtained for the case of 61 km sensing length, which was much better than that with traditional DMZIs [18]. Faraday rotating mirrors were used in the work of Ref. [19] for eliminating the effect of polarization induced fading in the experiment of 100-km sensing distance with a locating error of ±25 m. A cross correlation method was also used in the work to find the time delay between two signals.
In this paper, a new type of DMZI system is implemented for intrusion detection without using the aforementioned polarization control methods to eliminate the effect of polarization fading, and meanwhile a Fourier spectral analysis (FSA) method is proposed to locate the disturbance. Also, we have shown that the new type of DMZI system proposed here can always provide two identical signals that correspond to the clockwise and counter-clockwise waves, respectively.
The paper is organized as follows. Section 2 gives an outline of the experimental setup and describes the detected signals for the clockwise and counter-clockwise waves. In this section, the FSA method for locating the disturbance is also described, and the reliability of the presented system is stated. There, experimental results for a 250 meter and a 1036 meter long sensing fibers are presented, respectively. Section 3 introduces an average method to improve the locating accuracy. Then section 4 concludes this work.
2. Experiment and determination of location of intrusion
2.1. Experimental setup
The experimental system for disturbance detection and location is shown in Fig. 1, where a 250-meter- or 1036-meter-long fiber cable comprising four single-mode fibers (SMFs) is used as a disturbance sensor. Two of the single-mode fibers are for interference arms (i.e., sensing arms) of the MZI, while one of the other two fibers is used only for optical power delivery. Two polarization maintaining polarization beam splitters (PMPBS; named as PMPBS1 and PMPBS2) are used for splitting or combining optical power. When used as a splitting component, they split light into two orthogonally polarized lights for the two sensing arms of the MZI, respectively. When used reciprocally, they combine the two orthogonally polarized lights from respective arms into one polarization-maintaining (PM) fiber at the output end. Both PMPBSs are spliced with the sensing fibers to form a basic optical circuit for the modified DMZI system. Here, we assume PMPBS1 produces an x-polarized wave for the upper sensing arm of the MZI and a y-polarized wave for the lower sensing arm. The polarization maintaining fiber couplers (PMFC1 and PMFC2) used here work for only one polarized wave, say the x-polarized wave. That is, only the x-polarized wave can be split into two x-polarized lights, and reciprocally only two x-polarized waves can combine into one PM fiber. An in-line fiber polarizer is also used in the operation of the modified DMZI system. Within the areas framed by dashed boxes, as denoted by “PM component area” in Fig. 1, fiber components are all PM fiber types, and their lead fibers are all PM fibers. The symbol x in the PM component area represents a splice point between two fusion-spliced PM fibers at an offset angle of 45 degrees relative to each other’s principal axes.
Fig. 1. Dual Mach-Zehnder interferometer sensor system used in this study. FC: 50:50 fiber coupler, PD1 & PD2 : photodetectors, DAQ: data acquisition card, PMFC1 & PMFC2: polarization maintaining fiber couplers, PMPBS1 & PMPBS2: polarization maintaining polarization beam splitters, Φx/Φy: disturbance-induced phase variations for upper/lower arms at position d from the left end of the cable, L: length of the fiber cable. The symbol x represents a splice point between two spliced polarization maintaining fibers at an offset angle of 45 degrees relative to the principal axes of each other.The two PM component areas contain fiber components made from polarization maintaining fibers.
Laser light at 1550.12 nm is split into two laser beams through a 3 dB fiber coupler (FC) after boosted in power by an erbium-doped fiber amplifier. The two beams serve as light sources for the clockwise MZI and the counter-clockwise MZI, respectively. For the clockwise MZI, one laser beam passes through PMFC1, PMPBS1, sensing arms of the MZI, PMPBS2, in-line polarizer, power delivering fiber of the cable, and PMFC2, and is then detected by photodetector PD2. For the counter-clockwise MZI, another laser beam passes through PMFC2, power delivering fiber of the cable, in-line polarizer, PMPBS2, sensing arms of the MZI, PMPBS1, and PMFC1, and is then detected by photodetector PD1.
The signals received by both PD1 and PD2 are inputted to a data acquisition card (DAQ) taking 2M samples per second, and then analyzed by a personal computer (PC). Both time-domain and frequency-domain signals are used in the intrusion detection analysis.
2.2. Intrusion detection principle
To determine an intrusion, we constantly analyze the signal received by either PD1 or PD2. Three parameters derived from the received signal are : (1) SA (signal amplitude) defined as (Vmax–Vmin)/Vavg, where Vmax/Vmin/Vavg are maximum/minimum/average signal amplitudes of the time-domain signal waveform within the acquisition time period, which is 10 msec in this study; (2) FR (frequency ratio) defined by
When the fiber cable is heavily perturbed, SA might approach a large value in contrast to the case when the fiber cable is just lightly vibrated. Therefore, SA derived from the detected signal would be larger than a critical value (i.e., a threshold value) if an intrusion is determined. Similarly, when the fiber is heavily perturbed, the detected signal would oscillate fast, leading to a large LC compared to the case of lightly vibrating the fiber. A threshold for LC is thus needed for an intrusion to be determined as well.
On the other hand, a real intrusion would cause stronger high-frequency components in the spectrum of the detected signal than nuisance sources would. Thus choosing a larger value of υlow to cover a little higher frequency band for the numerator in Eq. (1) can rule out the effect of nuisance sources. Therefore, a man-made intrusion would give a larger FR than nuisance sources would. That is, FR needs to be larger than a threshold value for an intrusion to be determined. In this study, we chose υmin=100 Hz, υmax=5000 Hz, υlow = 2500 Hz, and υhigh=5000 Hz.
To determine the threshold values for SA, LC and FR, we have to find the ranges in which the three parameters fall, respectively, for non-intrusion cases and intrusion cases. To simulate the intrusion cases, we heavily perturbed the sensing fibers by letting an iron ball of 400 grams fall from a height of 20 cm and strike a section of fiber cable that was mounted on a rack (see Fig. 2 for a striking point at the position of L–d = 249 m). These cases were referred to as “heavily striking the cable” and corresponded to intrusion. Non-intrusion cases were simulated by lightly vibrating the fiber cable. Figure 3 shows the values of SA, LC and FR when the fiber cable was lightly vibrated 626 times (upper row), and when the fiber cable was heavily struck 526 times (lower row). Note that the sample numbers count from 1 to 626 (upper row), and from 1 to 526 (lower row), and each number represents a test sample. The value of SA rarely exceeded 2.0 when the fiber was lightly vibrated, while SA was larger than 2.0 most often in the case of intrusion. Therefore, we chose 2.0 to be the threshold for SA. For LC and FR, we chose 12 and 0.15 to be their thresholds, respectively, because lightly vibrating the fiber cable would make LC and FR exceed the thresholds only a few times, while heavily striking the fiber cable produced values of LC and FR larger than their respective thresholds most often. These statements were valid for signals detected by both clockwise and counter-clockwise photodetectors.
Fig. 2. A section of sensing fiber attached on a rack showing the position of L–d = 249 m where the fiber cable is struck.
Fig. 3. Values of SA, LC and FR when the fiber cable was lightly vibrated 626 times (upper row), and when the fiber cable was heavily struck 526 times (lower row).
Once these thresholds are predetermined this way, a real intrusion could then be determined if SA, LC and FR all simultaneously exceed their respective thresholds. From the results shown in Fig. 3, SA, LC and FR could exceed simultaneously their predetermined thresholds (indicated by the horizontal dashed lines) for most test samples in the case of heavily striking the fiber cable. Such a case will be discussed later again. However, the statistics from the 626 test samples in the case of lightly vibrating the fiber cable showed the following: (1) there was only one test sample with SA and FR exceeding their respective thresholds simultaneously, (2) LC and FR simultaneously exceeded their respective thresholds in only one test sample, (3) there was no test sample with SA and LC simultaneously exceeding the thresholds, and (4) none of the overall test samples gave values of SA, LC and FR that were simultaneously larger than their respective thresholds. Thus, none of these 626 tests would be determined as intrusion.
On the other hand, the test results for the case of heavily striking the fiber cable (the lower row of Fig. 3) showed the statistics that SA and LC would simultaneously exceed the thresholds in only 24 test samples, while SA and FR would simultaneously exceed the thresholds in only 23 test samples, and LC and FR would simultaneously exceed the thresholds in only 35 test samples. Besides, the three values of SA, LC and FR simultaneously exceeded their thresholds in 326 test samples out of the 526. That is, the other 200 test samples were not the case of intrusion because SA, LC and FR could not simultaneously exceed their respective thresholds. Note that we let the 400-gram iron ball fall on the fiber cable time after time till we completed collection of test data over a time period of <6 seconds, which gave 526 test samples each with a pair of time-domain waveforms of 10 msec. Here, “a pair” refers to the two signal waveforms detected by PD1 and PD2, respectively. Some of these test samples were acquired when the fiber cable was heavily perturbed, while some were acquired when the cable was just moderately (or even lightly) perturbed. In the latter case, SA, LC and FR might not simultaneously exceed the corresponding thresholds because some of their values were below the dashed line (see the lower row of Fig. 3). This is similar to the case when an intruder climbs the fence, producing heavy perturbation on the fiber most of the time and moderate perturbation at other times.
2.3. Principle of locating disturbance
2.3.1. Formulation of interference signals
When the fiber cable is disturbed at the position d from the left end of the interferometer, there induce phase variations Φx(t) and Φy(t) on the upper and lower sensing arms, respectively. Since PMFC1 and PMPBS1 are spliced at 45° relative to each other’s principal axes, the optical powers delivered to both sensing arms are equal, but would become different due to polarization variation incurred on the sensing arms (SMF), and therefore the waves at the input ends of PMPBS2 can be written as
Therefore only optical power changes at the output end of the in-line polarizer for the clockwise wave propagation. The horizontally polarized light at the output of the in-line polarizer would interact with its orthogonal counterpart before it reaches the PMFC2. However, since the PMFC2 only works for this horizontally polarized light, the optical power received by PD2 would vary. This power can be expressed as
For the counter-clockwise wave propagation, the x-polarized light out of the PMFC2 travels through the power delivering fiber of the cable, and reaches the in-line polarizer. Note that even the light may change its state of polarization after passing through the single-mode fiber or even may contain two slowly time-varying othorgonally polarized lights at the input end of the in-line polarizer, only x-polarized wave (i.e., the previously-mentioned horizontally polarized component for the clockwise-propagating wave) will travel to the 45° offset splice joint. After the offset splice joint, the wave can be expressed as
The two field components in Eq.(5) are then transmitted to the upper and the lower arms of the MZI, respectively. After traversing the sensing arms, the two waves can be expressed, respectively, as
Note that the ac components of the detected signals in Eqs. (4) and (8) are consistent with those reported in [18] except the noise terms from laser’s frequency noise, circuit noise, and interference between the signal light and Rayleigh backsattered light. In Eqs. (4) and (8), the variables A, B, C and D all vary with time due to polarization variation for the waves propagating down the sensing fibers and the power delivering fiber. However, these variables are slowly time-varying and their variations would not be appreciably observed within a short time period of 10 msec, which is the acquistion time period in the study. Then, they can be assumed to be constant within the acquisition time period. In addition, with the time-varying parts of the phases $\theta \; $and $\theta ^{\prime}$ in Eqs. (4) and (8) ignored, these phases just represent accumulative propagation phase delays and are equal. In this case, Eqs. (4) and (8) reflect a time delay of 2τ2 between the two detected signals, from which the disturbance position d can be known.
2.3.2. Fourier method for locating the intrusion
To determine the time delay 2τ2 and d (which is equal to L-cτ2), we do not use the conventional cross-correlation method, and use a Fourier spectral analysis (FSA) method instead. We take the Fourier transform of the signal waveforms received within every 10 msec time period. Fourier transforms of IPD1 and IPD2 are shown in Eqs. (9) and (10), where $|{{F_{PD1}}(\omega )} |$ and $|{{F_{PD2}}(\omega )} |$ are the corresponding Fourier amplitudes.
Fig. 4. Signal waveforms received by PD1 and PD2 (left), and their Fourier amplitudes (right) for a non-intrusion case.
Fig. 5. Signal waveforms received by PD1 and PD2 (left), and their Fourier amplitudes (right) for an intrusion case.
Figure 6 shows the signal waveforms received by PD1 and PD2(upper left), the corresponding Fourier amplitudes (lower left), the spectral dependence of computed position numbers over the frequency band from 100 to 5000 Hz (upper right), and the magnified drawing for the position numbers from 2500 to 5000 Hz (lower right) for an intrusion case in which the fiber cable is heavily perturbed at the position of d = 1 m (with L=250 m). Here the position number is defined as (L-d)/50, with L and d both expressed in meter and 50 being the spatial resolution in meter resulting from the sampling rate of the DAQ used in the experiment. The position number for each frequency in Fig. 6 was obtained by calculating the spectral phase difference 2ωτ2 at any given frequency ω and then deriving the value of L-d from the definition τ2 = (L–d) /c at each frequency. More specifically, the spectral phase difference obtained through the Fourier transforms in Eqs. (9) and (10), denoted here byφ,is related to the position number byφ = 3.0369 × 10−6 f P, where f = ω/2π is the spectral frequency and P the position number. In the study, the position number P is calculated by using P=3.2929 × 105φ/ f once the phase difference is obtained for each frequency. The position number represents a dimensionless measure of the distance from the end of the Mach-Zehnder interferometer. In the present case, SA, LC and FR are 2.48, 18 and 0.258, respectively, all exceeding the corresponding thresholds. Theoretically, the calculated spectrally-dependent time delay 2τ2 should correspond to a position number of 4.98 at every frequency ω. As we can see, however, from the calculated position numbers, the values of position number at many frequencies quite deviate from 4.98.
Fig. 6. Signal waveforms received by PD1 and PD2 (upper left), corresponding Fourier amplitudes (lower left), spectral dependence of computed position numbers over the frequency band from 100 to 5000 Hz (upper right), and magnified drawing for the position numbers from 2500 to 5000 Hz (lower right) for an intrusion case. Three frequencies, 3300 Hz, 3900 Hz and 4200 Hz, are marked (lower left) to emphasize weak spectral components, and corresponding unreasonable values of position number (lower right).
For frequencies smaller than 2500 Hz, the deviation is more prominent than those larger than 2500 Hz. And therefore, we only focus on those position numbers over the frequency band of 2500 to 5000 Hz by setting all position numbers below 2500 Hz to be zero (lower right, Fig. 6). Note that a positive position number corresponds to the the case that the waveform of IPD1 lags behind that of IPD2 while a negative position number corresponds to the opposite case. In the present case, position numbers should be always positive and their deviating too much from 4.98 results from the unequal time-varying parts of the accumulative phase $\theta \; $and $\theta ^{\prime}.$ To elliminate the position numbers with a negative value or too large magnitude, we consider only those frequency components having a strong enough amplitude in calculation. Here, we observe from Fig. 6 that three spectral components highlighted by circling at 3300 Hz, 3900 Hz and 4200 Hz are comparatively weak. To ignore these components and the like, we have defined the parameter “Characteristic Power (CP)” to be a ratio of a given spectral amplitude to the summation of the spectral amplitudes over the frequency band of 2500 to 5000 Hz. Any spectral component with a CP value smaller than some specific value, which is named “Characteristic Power Threshold (CPth)” here, will be ignored, and its position number will be nullified, i.e., set to be zero. Weak spectral components that have a CP value smaller than the threshold CPth are susceptible to noises occurring in photodetection and will incur incorrect position numbers. We have chosen 0.04 to be the value of CPth for years because such a choice has proven to be able to achieve low locating errors [20]. After averaging all the position numbers at the frequencies where their spectral components have a CP value ≥ 0.04, we then obtain the determined position number. In the case of Fig. 6, the determined position number is 5.344. The discrepancy between the calculated and the true position numbers is 0.364, which corresponds to a locating error of 18.2 m (=0.364 × 50 m).
For another case of intrusion, we show, in Fig. 7, the signal waveforms received by PD1 and PD2, the corresponding Fourier amplitudes, and the spectra of phase differences and position numbers over the frequency band from 100 to 5000 Hz. In this case, a 1036 m long fiber cable was heavily perturbed at the position of d = 787 m, which corresponded to the position number 4.98. Note that the spectral phase differences were obtained first by deriving the phase difference between the Fourier components of IPD1 and IPD2, and then calculating the position number for each frequency by the P-φ relation above. The determined position number for this case was 5.137 when CPth=0.04 was used.
Fig. 7. Signal waveforms received by PD1 and PD2 (upper left), corresponding Fourier amplitudes (lower left), spectra of phase differences (upper right) and position numbers (lower right) over the frequency band from 100 to 5000 Hz and u) for an intrusion case in which a 1036 m long fiber cable was heavily perturbed at the position of d=787 m (corresponding to a position number of 4.98).
2.4. Reliability of the DMZI system
The DMZI system for intrusion detection and locating has been demonstrated in our lab since 2016 [21]. A DFB laser with a 20 MHz linewidth and with its power boosted by an erbium-doped fiber amplifier was used as a light source (43 mW) in the system first developed in 2016. Figure 8 shows a pair of signal waveforms detected, respectively, by PD1 and PD2 for the cases of non-intrusion (upper left) and intrusion (lower left) when the fiber cable was perturbed at the position of d = 145 m (corresponding to the position number 2.1) [20]. The corresponding Fourier amplitudes for both cases are also shown in the figure. This experiment was conducted in 2017 when we started to discuss the effect of CPth, following the initial work in 2016. At that time, the thresholds for the parameters SA, LC and FR were 1.1, 12 and 0.1, respectively. Intrusion determination was based on whether the three parameters simultaneously exceeded their respective thresholds. In both cases, oscillation patterns of the two signals detected by PD1 and PD2 were quite similar, however with the spectral content having more pronounced high frequency components in the case of intrusion.
Fig. 8. Signal waveforms received by PD1 and PD2, and corresponding Fourier amplitudes, for a non-intrusion case (upper row) and an intrusion case (lower row).
The signal waveforms for the case of intrusion shown in Fig. 8 were processed to give the frequency-dependent position numbers over 2500 to 5000 Hz as indicated in Fig. 9, where the position numbers from 3200 Hz to 5000 Hz were cleared to null because the spectral components of the signal in this frequency range did not have a CP value reaching 0.04 in this case. The average value of these nonzero position numbers is 2.401, which accounts for a discrepancy of 0.301 from the true value of 2.1, and such discrepancy is equivalent to a spatial error of 15.05 m in locating the disturbance/intruder.
Fig. 9. Position numbers calculated over the frequency band from 2500 to 5000 Hz for the case of intrusion in Fig. 8. Here CPth=0.04 was used to nullify the position numbers for weak spectral components.
In 2018, studies of the same DMZI system continued with only a change in the laser source type being made [22]. An erbium-doped fiber laser of a ring type was constructed to have a linewidth of 20 KHz and an optical power of 25 mW (after boosted by an erbium-doped fiber amplifier), and has been used for subsequent studies since. Results shown in Figs. 3 to 6 were obtained using this system in 2018. We can see that the interferometric visibilities for both detected signals were almost equal to unity as the highly coherent laser was used. However, the use of a DFB laser could produce the same level of locating accuracy even with a lower visibility (see Figs. 8 and 9). In 2019, the length of the sensing interferometer was extended to 1036 m by splicing three single-mode fibers of an equal length of 786 m to, respectively, the three fibers (two sensing fibers and one power delivering fiber) of the existing fiber cable [23], with all other optical components unchanged. Then, we heavily perturbed the fiber at the position of d=393 m (corresponding to the position number 12.86) by letting an iron ball of 400 grams fall from a height of 20 cm and hit a suspended thin metal plate on which parts of sensing fibers were attached. In an example, the signal waveforms detected by PD1 and PD2 as well as the corresponding Fourier spectral amplitudes are shown in Fig. 10. The calculated SA, LC and FR for IPD2 signal were, respectively, 2.662, 20 and 0.219, all exceeding their respective thresholds. Note that SA, LC and FR were almost the same for IPD1 and IPD2 signals. The inset in the Fourier amplitude spectrum represents the position number versus frequency as CPth=0.04 was used to clear the contribution from weak spectral components. The average of these non-zero position numbers was then found to be 13.173, which represented a locating error of 15.65 m.
Fig. 10. Signal waveforms received by PD1 and PD2, and corresponding Fourier amplitudes, for an intrusion case (with the position number of 12.86 or d=393 m). The results were obtained in 2019.
With the system configuration unmodified and the thresholds for the three parameters unaltered, the DMZI system with a 1036 m sensing length has been studied in these two years aiming to pursuit more accurate results in locating the disturbance. Parts of the recent research works deal with different methods from this work for determining the location of the intrusion, and relevant results are not presented here. Instead, we show signal waveforms received by PD1 and PD2 for the case of heavily perturbing the fiber at three positions, i.e., d=1035 m, d=787 m, and d=393 m (corresponding to the position numbers 0.02, 4.98, and 12.86, respectively) in Fig. 11, where the corresponding position numbers for CP values not smaller than CPth=0.04 from 100 Hz to 5000 Hz are also shown. The estimated position number for each case is then the average of these non-zero position numbers. From top to bottom, the estimated position numbers are, respectively, 0.405, 4.697, and 13.024. The discrepancies between the estimated and real locations of intrusion are then 19.25 m, 14.15 m, and 8.2 m, respectively. The algorithm for locating the disturbace still follows the Fourier spectral analysis method that has been used for years. We show these results here just to demonstrate the applicability and reliability of the presented DMZI system without changing any system components and the thresholds for intrusion determination. The presented DMZI system has maintained the same level of performance for such a long time. At the present time, the DMZI system is still working to give comparable capability of locating the intrusion.
Fig. 11. Signal waveforms received by PD1 and PD2, and position number versus frequency for the case of heavily perturbing the fiber cable, respectively, at the positions of d = 1035 m, d=787 m, and d=393 m (from top to bottom). Here CPth=0.04 was used to nullify the weak spectral components. The results were obtained in 2020 [24].
The results shown above were all based on signals acquired within a time period of 10 msec. A sudden phase change within a short time period might bring about large SA, LC and FR that all exceed their respective thresholds, and an intrusion is determined in this case. However, this case ought not to be considered as an intrusion sometimes. To explain further, we take a non-intrusion case in which a sudden vibration of the fiber cable lasts for only 0.3 second. This situation may occur when a bird suddenly takes off from the fiber cable or when hailstones fall upon the fiber cable. Some high frequency components may continuously appear in the Fourier spectrum of the time-domain signal. Because each acquisition time period is only 10 msec, there might be consecutively 30 time-domain signal waveforms that produce large SA, LC and FR. However, in most of subsequent acquisition time periods, SA, LC and FR are small. If we only examine the 30 signal waveforms or any of them, we might determine an intrusion in this situation, in which an intrusion is not supposed to be detected though. To avoid this misdetermination, a large number of sequentially-acquired signal waveforms are better to be examined. If most signal waveforms acquired every 10 msec within a longer time period (one second for example) are all determined as an intrusion, an intrusion event can be claimed. Then in the previous case of sudden phase variation, no intrusion would be determined, and this is correct. On the contrary, the phase variation would continue for a longer time when someone climbs a fence, and the case can be determined to be an intrusion event. To derive the location of intrusion, an average method is thus used, as explained in the following section.
3. Average method toward practical detection and location
Experiments carried out in the past 5 years have demonstrated good locating accuracy in most examples in which signal waveforms are acquired on a basis of 10 msec span. Specifically speaking, most signal waveforms acquired could be reduced to locating errors smaller than 16 m, but a small number of signal waveforms could be reduced to larger locating errors. An extreme and rare case is shown in Fig. 12, where a pair of signal waveforms detected by PD1 and PD2 (left) and frequency dependence of the position number (right) are shown for the case of heavily perturbing the fiber cable (with a length of 250 m) at the position of d=1 m (corresponding to the position number 4.98). In this case, SA, LC and FR for IPD1 signal were, respectively, 2.251, 20 and 0.196, all exceeding their own thresholds. Although the two signal waveforms quite matched with each other except with a time delay, the position numbers shown on the right hand side of the figure exhibited quite large values at some frequencies such as 2600 Hz, 3000 Hz and 3100 Hz. The value of CPth was set at 0.04 to nullify the position numbers of the weak spectral components over the band of 2500 Hz to 5000 Hz in this case. The resultant position number was then calculated to be 9.742 by taking an average of these non-zero position numbers over this band. The calculated position number showed a discrepancy of 4.762 relative to the true value, and corresponded to a locating error of 238.1 m. Sometimes, locating errors between 100 and 200 m could occur due to detection noises and environmental disturbances on the fiber at other locations. However, we could still maintain a good enough locating accuracy by using an average method in which a series of detected signals within every longer time period, say one second, were examined.
Fig. 12. Signal waveforms received by PD1 and PD2, and position number versus frequency for the case of heavily perturbing the fiber cable at the position of d=1 m (corresponding to the position number 4.98). Here CPth=0.04 was used to nullify the contribution from weak spectral components.
When the fiber cable was heavily perturbed at the same position many times, we observed that the locating error ranged from > 0 m to >238 m, however with 81.9% of locating errors being smaller than 100 m, with 69.3% of locating errors being smaller than 75 m, with 51.2% of locating errors being smaller than 50 m, and with 26.1% being smaller than 25 m.
Therefore the location of intrusion can not be determined by merely examining the received signals acquired every 10 msec time period, because the locating error may vary within a large range for a given intrusion location. From a practical viewpoint of intruders’ action on a netted fence, the time consumed to complete climbing, breaking, or cutting is at least one second. Also, when a ball or a stone is maliciously thrown at the fence, the fiber cable would be heavily vibrated for half a second or longer. That is, the determination of an intrusion and the locating of the intrusion can be done by using 100 pairs of signal waveforms with each pair acquired in a 10 msec time period. In every one second, there are 100 pairs of signal waveforms (with each pair referring to the signal waveforms received by PD1 and PD2) acquired. Each event is then defined by what is happening in the time period of every 100 pairs of signal waveforms. For an intrusion event, the values of SA, LC and FR for most pairs (or at least 50% in quantity) of signal waveforms are required to exceed their respective thresholds. As previously mentioned, the locating errors derived from a large number of signal waveforms distribute in a range, but most of the locating errors are smaller than 50 m. To verify this, we first heavily perturbed the fiber cable of 250 m in length at the position of d=1 m (corresponding to the position number 4.98) for 5 seconds, acquiring 500 pairs of signal waveforms. These signal waveforms were then processed in the sequence of events. That is, we examined the 100 pairs of signal waveforms acquired sequentially. For each 100 pairs, we checked if every pair of signal waveforms corresponded to an intrusion case by examining SA, LC and FR. If most pairs were reduced to an intrusion case, then the 100 pairs as a whole corresponed to an intrusion event. In accodance with the previous caculation, the location of intrusion would be determined by those pairs (58 pairs for example) that corresponded to an intrusion. By averaging these calculated locations of intrusion (e.g.,58 calculated locations of intrusion), we then obtained the location of intrusion for the event.
Table 1 shows the average location of intrusion determined from each 100 pairs of detected signal waveforms. Here, we can see how many pairs of detected signal waveforms corresponded to an intrusion case for every 100 pairs or in every one second. Because the fiber cable was perturbed heavily in a continous way for five seconds, the system determined 5 intrusion events because the number of pairs that were reduced to intrusion reached the majority in each event. The discrepancy between the calculated location of intrusion and the real location was no larger than 0.3078 in position number, which represented a maximum locating error of 15.39 m (with respect to the minimum locating error of 1.015 m for the fifth intrusion event).
Table 1. Average location of intrusion (expressed in position number) for five intrusion events occuring at the position of d=1 m. Sequentially, 100 pairs of detected signal waveforms were acquired and the number of pairs that were reduced to intrusion was calculated for each event.
We then heavily perturbed the fiber cable at the position of d=249 m (corresponding to the position number 0.02) again for 5 seconds, acquiring 500 pairs of signal waveforms. Table 2 shows the average location of intrusion determined from each 100 pairs of detected signal waveforms. Here again, we can see how many pairs of detected signal waveforms corresponded to an intrusion case for every 100 pairs or in every one second. Because the number of pairs that corresponed to intrusion reached the majority, each sequence of 100 pairs was then considered to be an intrusion event. The discrepancy between the calculated location of intrusion and the real location was no larger than 0.5155 in position number, which represented a maximum locating error of 25.775 m (with respect to the minimum locating error of 12.345 m for the fourth intrusion event).
Table 2. Average location of intrusion (expressed in position number) for five intrusion events occuring at the position of d=249 m. Sequentially, 100 pairs of detected signal waveforms were acquired and the number of pairs that were reduced to intrusion was calculated for each event.
Three pairs of the detected signal waveforms were chosen from the 500 pairs in the case of Table 1 and are shown in Fig. 13. These three examples all corresponded to intrusion, giving locations of intrusion of 4.8101, 4.3625, and 4.8335 (from top to bottom), respectively. Also, three pairs of the detected signal waveforms were chosen from the 500 pairs in the case of Table 2 and are shown in Fig. 14. These three examples also represented cases of intrusion, and the calculated locations of intrusion were -0.2222, -0.2676 and 0.1252 (from top to bottom), respectively. In these six examples, the largest locating error was 30.875 m, but all of the others were smaller than 14.4 m.
Fig. 13. Signal waveforms received by PD1 and PD2, and position number versus frequency for the case of heavily perturbing the fiber cable at the position of d=1 m (corresponding to the position number 4.98). Here CPth=0.04 was used to nullify the weak spectral components.
Fig. 14. Signal waveforms received by PD1 and PD2, and position number versus frequency for the case of heavily perturbing the fiber cable at the position of d=249 m (corresponding to the position number 0.02). Here CPth=0.04 was used to nullify the weak spectral components.
4. Conclusion
A new DMZI system aiming to detect and locate the intrusion is proposed and tested. In this work, we have presented the theoretical background for the DMZI sensor that can maintain high interference between the two sensing arms without using a polarization control loop. Both clockwise and counter-clockwise interference signals are shown to be highly similar to each other with a time delay. During the test period from 2016 to 2020, we changed only two things regarding the operation of the optical system. First, the laser source type was switched in 2018 from a DFB laser of 20 MHz linewidth to a fiber laser of 20 KHz linewidth. Secondly, the length of the sensing interferometer was extended in 2019 from 250 m to 1036 m. The detected signal waveforms exhibited high interference visibilities for both clockwise and counter-clockwise waves, indicating that there was no polarization effect on this new system.
We have also presented a Fourier spectral analysis method to determine the location of intrusion by examining the phase difference between each pair of spectral components of the detected signals. After ignoring the position numbers contributed by weak spectral components, we then average all position numbers in a given frequency band. This band is chosen to contain high frequency spectral components, such as those from 2500 Hz to 5000 Hz, because high frequency signals, if strong enough to combat against noises, can give better resolution in determining the time delay between the two signals. Experimental results have demonstrated the reliability of the proposed DMZI system with the Fourier spectral analysis method used for locating the intrusion. We have shown that the locating error could be smaller than 50 m when the fiber cable was heavily perturbed by letting a iron ball of 400 grams fall from a height of 20 cm and repeatedly strike the fiber cable attached on a rack for several minutes. Many examples proved that the signal waveforms detected by PD1 and PD2 could be reduced to locating errors of <20 m. These results were reported in these five years based on singal waveforms detected by PD1 and PD2 in a time span of 10 msec. However, the odds were that such 10-msec signal waveforms would be reduced to locating errors in the order of 100 m. The reason for such large locating errors could be attributed to electronic noises added to the signal waveforms and the disturbances induced on the fiber cable at other locations. To resolve this problem, we propose a method to detect an intrusion and determine the location of intrusion using 100 pairs of sequential signal waveforms. In the method, each pair of signal waveforms are examined to see if they meet the intrusion condition. Then, the number of the reported intrusions in the 100 pairs is counted within the time period of one second. If the number exceeds 50% of the total at least, an intrusion is claimed. The locations of intrusion for these pairs of signal waveforms that correspond to intrusion are then calculated in accordance with the algorithm presented. An average of these locations of intrusion gives a location of intrusion for the event of concern in a one-second time period. Experimental results have shown that the locations of intrusion for five intrusion events occurring at the position of d=1 m can be determined with the spatial errors of 1.015 m to 15.39 m, while five intrusion events occurring at the position of d=249 m are located accurately to within 12.345 m to 25.775 m.
In conclusion, the modified DMZI sensor system with new locating algorithm proves to be a simple and robust disturbance-sensing system. Experimental results obtained from 2016 up to now have demonstrated the capability of intrusion detection with a low locating error (with an maximum error of 25.775 m) by taking 100 pairs of signal waveforms into consideration for intrusion determination.
Funding
Ministry of Science and Technology, Taiwan (109-2221-E-007-107).
Acknowledgements
The authors acknowledge Prof. Tai-Lang Jong, National Tsing Hua University for his help in preparing the data acquiring program several years ago.
Disclosures
The authors declare that there are no conflicts of interest related to this article.
Data availability
Most data underlying the results presented in this paper are available in. References [20] to [24]. Some data of the results presented are not publicly available at this time, but may be obtained from the authors upon reasonable request.
References
1. J. C. Juarez, E. W. Maier, K. N. Choi, and H. F. Taylor, “Distributed fiber-optic intrusion sensor system,” J. Lightwave Technol. 23(6), 2081–2087 (2005). [CrossRef]
2. Y. Lu, T. Zhu, L. Chen, and X. Bao, “Distributed vibration sensor based on coherent detection of phase-OTDR,” J. Lightwave Technol. 28(22), 3243–3249 (2010). [CrossRef]
3. A. Masoudi, M. Belal, and T. P. Newson, “” A distributed optical fibre dynamic strain sensor based on phase-OTDR,”,” Meas. Sci. Technol. 24(8), 085204 (2013). [CrossRef]
4. F. Peng, H. Wu, X.-H. Jia, Y.-J. Rao, Z.-N. Wang, and Z.-P. Peng, “” Ultra-long high- sensitivity Φ-OTDR for high spatial resolution intrusion detection of pipelines,”,” Opt. Express 22(11), 13804–13810 (2014). [CrossRef]
5. M. Aktas, T. Akgun, M. U. Demircin, and D. Buyukaydin, “Deep learning based multi-threat classification for phase-OTDR fiber optic distributed acoustic sensing applications,” Proc. SPIE 10208, 102080G (2017). [CrossRef]
6. J. P. Dakin, D. A. J. Pearce, A. P. Strong, and C. A. Wade, “A novel distributed fiber sensing system enabling location of disturbances in a Sagnac loop interferometer,” Proc. SPIE 0838, 325–328 (1988). [CrossRef]
7. X. Fang, “A variable-loop Sagnac interferometer for distributed impact sensing,” J. Lightwave Technol. 14(10), 2250–2254 (1996). [CrossRef]
8. W. Xu, C. Zhang, S. Liang, L. Li, W. Lin, and Y. Yang, “Fiber-optic distributed sensor based on a Sagnac interferometer with a time delay loop for detecting time varying disturbance,” Microwave Opt. Technol. Lett. 51(11), 2564–2567 (2009). [CrossRef]
9. X. Hong, J. Wu, C. Zuo, F. Liu, H. Guo, and K. Xu, “Dual Michelson interferometers for distributed vibration detection,” Appl. Opt. 50(22), 4333–4338 (2011). [CrossRef]
10. Q. Li, H. Wang, L. Li, S. Liang, and X. Zhong, “Fiber-optic sensor based on Michelson interferometers for distributed disturbance detection,” Infra. Laser Eng. 44(1), 205–209 (2015).
11. S. J. Spammer, P. L. Swart, and A. A. Chtcherbakov, “Merged Sagnac–Michelson interferometer for distributed disturbance detection,” J. Lightwave Technol. 15(6), 972–976 (1997). [CrossRef]
12. Q. Song, P. Zhou, H. Peng, Y. Hu, Q. Xiao, H. Wu, and B. Jia, “Improved localization algorithm for distributed fiber-optic sensor based on merged Michelson-Sagnac interferometer,” Opt. Express 28(5), 7207–7220 (2020). [CrossRef]
13. A. A. Chtcherbakov, P. L. Swart, and S. J. Spammer, “Mach–Zehnder and modified Sagnac-distributed fiber-optic impact sensor,” Appl. Opt. 37(16), 3432–3437 (1998). [CrossRef]
14. A. A. Chtcherbakov, P. L. Swart, S. J. Spammer, and B. M. Lacquet, “Modified Sagnac/Mach-Zehnder interferometer for distributed disturbance sensing,” Microwave and Opt. Technol. Lett. 20(1), 34–36 (1999). [CrossRef]
15. G. Luo, C. Zhang, L. Li, Z. Ma, T. Lan, C. Li, and W. Lin, “Distributed fiber optic perturbation locating sensor based on dual Mach-Zehnder interferometer,” Proc. SPIE 6622, 66220z (2008). [CrossRef]
16. D. Tu, S. Xie, Z. Jiang, and M. Zhang,”Ultra long distance distributed fiber-optic system for intrusion detection,” Proc. SPIE 8561, 856111 (2012). [CrossRef]
17. Q. Chen, T. Liu, K. Liu, J. Jiang, Z. Ding, L. Zhang, Y. Li, L. Pan, and C. Ma, “An elimination method of polarization-induced phase shift and fading in dual Mach–Zehnder interferometry disturbance sensing system,” J. Lightwave Technol. 31(19), 3135–3141 (2013). [CrossRef]
18. C. Ma, T. Liu, K. Liu, J. Jiang, Z. Ding, L. Pan, and M. Tian, “Long-range distributed fiber vibration sensor using an asymmetric dual Mach–Zehnder interferometers,” J. Lightwave Technol. 34(9), 2235–2239 (2016). [CrossRef]
19. J. Huang, Y. Chen, Q. Song, H. Peng, P. Zhou, Q. Xiao, and B. Jia, “Distributed fiber-optic sensor for location based on polarization-stabilized dual-Mach-Zehnder interferometer,” Opt. Express 28(17), 24820–24832 (2020). [CrossRef]
20. H.-R. Ho, Study of Fiber-Optic Distributed Intrusion Detection and Location Sensor Using Dual Mach-Zehnder Interferometers, Master thesis, Nat’l Tsing Hua Univ., Taiwan, R.O.C. (2017), Chap.4.
21. Y.-L. Gao, Study of Modified Dual Mach-Zehnder Interferometers for Distributed Vibration Sensor, Master thesis, Nat’l Tsing Hua Univ., Taiwan, R.O.C. (2016), Chap.4.
22. C.-Y. Hsieh, Study of Dual Mach-Zehnder Interferometer Intrusion Detection and Location Using Fiber Laser: Location Resolution Improvement by Averaging Method, Master thesis, Nat’l Tsing Hua Univ., Taiwan, R.O.C. (2018), Chap.4.
23. Y.-C. Hsu, Study of Intrusion Detection and Location Using Dual Mach-Zehnder Interferometers with Fiber Laser as Light Source: Locating Two Different Places of Intrusion at the Same Time, Master thesis, Nat’l Tsing Hua Univ., Taiwan, R.O.C. (2019), Chap.4.
24. B.-Y. Huang, Analysis of Intrusion Detection and Location Using Dual Mach-Zehnder Interferometers: Machine Learning Algorithms, Master thesis, Nat’l Tsing Hua Univ., Taiwan, R.O.C. (2020), Chap.4.
