1. Introduction

Fiber temperature and strain sensors based on Brillouin scattering (BS) were introduced in 1989 by Culverhouse et al. [1] and Horiguchi et al. [2] respectively. That very same year, Horiguchi and Tateda proposed a way of making these sensors fully distributed by using a pulsed pump laser [3] enhancing their attractiveness for various sensing applications. The two main techniques used for distributed BS fiber sensors are called Brillouin Optical Domain Reflectometry (BOTDR) [1] and Brillouin Optical Time Domain Analysis (BOTDA) [4]. Nowadays, other techniques such as Brillouin Optical Correlation Domain Analysis (BOCDA) [5] also exist, usually exhibiting a very high spatial resolution (mm scale), but have limited sensing reach (usually in the order of hundreds of meters) [6–8]. The spatial resolution of BOCDA system is determined by the modulation parameters of the light source such as the amplitude and frequency compared to BOTDR and BOTDA systems in which the resolution is based on the decay time of an acoustic wave. Since its discovery, many groups have worked on increasing the sensing distance achieving up to 150 km [9] as well as maximizing the spatial resolution down to 2cm in standard optical fiber [10] using conventional BOTDR/A systems. However, the sensitivity of ~1.1MHz/°C in temperature [11] and of ~0.05MHz/µε in strain [12] has remained mostly unchanged using the first spontaneous Stokes shift. In 2014, the authors demonstrated a technique that enabled the enhancement in sensitivity by a factor of n times the standard sensitivity of BS based sensors by using an nth order stimulated Brillouin scattering (SBS) Stokes wave [13]. We showed a linear increase with the order of Stokes waves up to 6-fold standard sensitivity. Similar results with a different ring configuration were subsequently demonstrated by Xu and Zhang with the 12th Stokes wave [14]. A 3-fold increase was also demonstrated by Liu et al. with a 10⁻⁶ °C resolution by operating the Stokes waves in single longitudinal mode regime [15]. However, these demonstrations were not on distributed sensing which limits their potential commercial applications as the temperature measurement is made over the entire integrated fiber length.

In this paper, we demonstrate a distributed temperature sensor scheme using the enhanced sensitivity from higher order SBS in optical fiber. We show the working principle of a fully distributed BOTDA-like sensor with a sensitivity of 4 MHz / °C and a spatial resolution around 225 m over more than 4 km of sensing fiber. Maximum sensing range was not the focus of this article, therefore no effort was made on optimization, but it could be easily increased since this setup is not length dependent in terms of SBS generation. The current limited resolution is due to equipment issues and noise, which could be potentially mitigated using better components. To our knowledge, this is the first working prototype for such a distributed sensor with higher sensitivity.

2. Theory

BS is a third order nonlinear effect which involves scattered Stokes wave from the interaction of a pump wave with acoustic phonons in a material. Typically, in optical fiber the loss of energy from the pump to the acoustic phonon results in a slightly red-shifted counter-propagating Stokes wave with a frequency shift of ~11 GHz. As more energy is transferred to the Stokes wave, the beating between the pump and the Stokes waves increases the number of acoustic phonons at the Brillouin frequency shift through electrostriction which results in more light being scattered. This phenomenon is known as stimulated Brillouin scattering. The Brillouin frequency shift, ${\nu }_B$is given by,

Where n_eff(T, ε) is the material refractive index which is dependent on temperature and strain, V_A(T, ε) is the phonon velocity and λ is the pump wavelength.

Temperature and strain influence the refractive index which we can make use of by using the BS frequency shift for sensing since the dependence is linear as shown in Eq. (2).

Where C_T is the temperature-frequency coefficient, C_ε is the strain-frequency coefficient and ν_B0 is the Brillouin frequency shift at a reference temperature T₀, and strain reference, ε₀. To detect the variation of the Brillouin frequency shift induced either by temperature or strain precisely, a heterodyne beat frequency technique is usually applied to produce a better resolution compared to using an optical spectrum analyzer (OSA) which has limited spectral resolution.

SBS lasers generating multiple Stokes shifts have been demonstrated in a wide range of designs [16–18]. Such systems reduce the SBS threshold by using a ring cavity configuration which commonly leads to generating cascaded stimulated Brillouin scattering (cSBS) and therefore to a frequency comb composed of high-order Stokes waves. To measure the increased sensitivity from high order Stokes waves, a beat signal from the mixing of reference SBS oscillator and SBS sensor oscillator signals is displayed on an electrical spectrum analyzer (ESA). If both the reference and sensing fibers are identical (ν_B0,ref = ν_B0,sens), the resulting beat-note signal is only related to the multiple order Stokes waves interfering with its respective reference oscillator Stokes order and is given by Eq. (3).

Where C_2n,T and C_2n,ε are the linear coefficients of the dependence of each high order Stokes beat signals on the sensed parameters (temperature and strain). For example, C_4,T will be twice as large as C_2,T and so on as schematically shown in Fig. 1.

 

Fig. 1 Schematic of the self-heterodyne detection scheme. Similar reference and sensing oscillators generate a cSBS frequency comb from the same laser source and are recombined at an ESA where they are analyzed. Difference in the SBS frequency combs leads to a beat frequency spectrum with multiple peaks. Variation in temperature or strain leads to a shift in the beat frequencies close to baseband related to the Stokes orders in the ESA.

Download Full Size | PDF

Figure 1 shows that the reference high-order Stokes frequencies of the spool at temperature, T₀ or strain, ε₀ beats with the shifted SBS Stokes frequencies of the sensing spool at T or ε. By mixing S_2,sens with S_2,ref, S_4,sens with S₄, ref and so on, discrete beat frequencies are observed at the ESA, but all are at the low base-band frequency. If both the reference and sensor spools are identical, all the beat frequencies between identical Stokes orders generate an overlapping DC signal. Any difference in temperature or strain will generate a cascade of beat frequencies that increase in their separation as a function of the Stokes order. Hence the nth Stokes order will shift n times faster than the spontaneous frequency shift. To avoid overlapping beat signals from different orders, different types of fibers can be used between reference and sensing fiber to cause a frequency shift offset.

A key feature describing SBS behavior is the power threshold required to generate a Stokes wave. Typically, shorter active fiber sections make it harder to generate SBS since the SBS threshold is inversely proportional to the length of the fiber, as described by the following relation for single-pass SBS generation:

where the number 21 is a numerical approximation reported for the first time by Smith in 1972 [19], A_eff is the effective area, L_eff is the effective length and is given by$L_{eff}=(1-e^{-\alpha L})/\alpha$ which takes into account the loss of the fiber and simplifies typical mode equation calculation. In the case of distributed sensing, the active length is short, therefore L_eff ≈L_active. The term in brackets is a factor that shows the dependence on the linewidth of the pump and the Stokes wave over the SBS power threshold which tends to 1 when the pump linewidth ∆ν_L is small. Due to recirculation of the light in a ring cavity, the threshold is lowered and is given by Eq. (5) [20].

where L is the ring cavity length, R_m is the fraction of Stokes power reinjected after each round trip. This represents the threshold for each Stokes wave. Therefore, a Stokes wave must reach this threshold for the subsequent Stokes to be generated, thus increasing the requirements of the seed laser. To circumvent such a requirement for increased power, internal linear gain using an amplifier can be applied in the cavity to compensate for loss (α and R_m) by increasing each Stokes wave power to the threshold level for the next order. However, such a gain must be controlled meticulously as it can also initiate lasing of the cavity, a common problem with multi-Stokes SBS cascaded lasers [21].

3. Experimental Setup

Two near identical Brillouin ring oscillators are used to monitor the evolution of the Stokes spectra; one as a reference and a second one, with integrated time-gating for distributed measurements, as the sensor with the fiber under test (FUT). Seed light, commonly known as the Brillouin pump (BP) is sent to both oscillators via a 3dB-coupler. The seed power can be controlled with an erbium doped fiber amplifier (EDFA). The sensor oscillator is described next for simplicity, but note that both oscillators have similar behavior, except that the reference cSBS laser is not time gated. This reference uses a cSBS oscillator seeded with the same BP as the sensing cavity instead of an independent single-frequency reference laser to avoid jitter-noise between the two sources.

Once light from the BP is separated in a 3dB-coupler, it enters each cavity through another 3dB coupler. As the BP power is increased, the 1st Stokes line is generated in the opposite direction and oscillates in the bottom branch of the oscillator through optical circulators as shown in Fig. 2. A cascaded process then occurs, as more BP power is injected, light is scattered into the 1st Stokes wave. At a certain point, this Stokes wave is intense enough and reaches the threshold for the 2nd Stokes wave, which oscillates in the upper part of the cavity e.g. same direction as the initial BP. Odd or Even Stokes waves can be collected at the output with a 99:1 output coupler. The in-cavity EDFAs are used to compensate for the loss and keep the Stokes wave lasing above threshold. An optical bandpass filter is used in this configuration to decrease the ASE bandwidth of the EDFAs to only a few nanometers (~1nm for the sensing cavity and 5nm for the reference oscillator) reducing the potential of free running modes competing for gain in such a configuration [21]. A 2.5 km length of fiber on a reel was placed in a temperature controlled chamber, while the total length of fiber under test (FUT) of the cavity was slightly above 4 km, giving a free spectral range (FSR) of 49.707 kHz.

 

Fig. 2 Two near identical SBS ring resonators are used; one as a reference and the other as the sensor, both sharing a common seed laser through a 3dB coupler. The signals are recombined at their respective output by a second 3dB coupler connected to the electrical spectrum analyzer. AOMs are used as temporal gates which provide the spatial resolution of the sensor. One AOM is electrically controlled to vary the time of the overlap with the other AOM to allow a scan over the entire length of the fiber spool. The in-cavity EDFAs are used to compensate for the cavity loss.

Download Full Size | PDF

Since the time position of the overlap of counter-propagating pulses is controllable by the relative delay between the acousto-optic modulators (AOMs), distributed temperature or strain sensing (DTS) is possible, i.e. cascaded stimulated Brillouin scattering (cSBS) occurs only in the pulse overlap region where the intra cavity gain overcomes the intra-cavity loss. The advantage of the AOM is polarization independence and low loss. A shorter gating time leads to better temporal resolution and shorter sensing length. However, reducing the resolving length, makes it harder to generate an SBS frequency comb. Figure 3 shows the time-gating DTS scheme. It is important to ensure that SBS is only generated in this overlapping time window, which implies that the cavity gain must be maintained below threshold when the AOMs are closed. Electrically delaying one AOM’s opening with respect to the other, changes the position of the overlap region in the fiber, therefore allows spatial discrimination and control. By performing a scan of the relative delay between the opening of the AOMs, the temperature variation along the entire length of the fiber can be monitored. The output light from the sensor is recombined with the reference oscillator kept at constant temperature and the beat spectrum displayed on an ESA, as shown schematically in Fig. 1. Since the beat frequencies are at the base-band, a simple oscilloscope could be used instead for such measurement. The difference in beat frequency can be related to temperature variations and as described by Eq. (3). The Stokes waves order, n, thus enhances the sensitivity by n fold over the BS scheme.

 

Fig. 3 Description of the influence of AOMs gate-overlap on the cSBS generation, depicted in grey (gate 1) and blue (gate 2) in the lower part of the figure. In a) and c) the AOM gates do not overlap and the cavity loss ensures SBS does not reach threshold. In b) the AOM gates do overlap and the gain in the region is high enough for SBS to be cascaded. Control of the opening of the AOMs temporally in opposite direction, allows overlap only in a certain region of the fiber which corresponds to the spatial resolution of the system (shorter temporal gate time means better resolution).

Download Full Size | PDF

Figures 3(a) and 3(c). shows the cases when the gates are out of synchronism in the cavity and hence, no Stokes waves are observed. In Fig. 3(b), the convolution product of the two gates provides sufficient power to enable cSBS generation. One gate is open for a very short time and the other is open for longer. This allow the convolution product to be sharper (more rectangular) than with two identical timing gates (triangular convolution function). Finally, the resolution is defined by the smallest cross product of the two gates.

It is important to note that SBS generated in the overlap window is only possible after multiple passes. In other words, the sensing dual-cavity is effectively a mode-locked laser, where the Stokes wave used as a probe must be lasing (gain>loss). The modulating frequency on the AOMs must therefore be equal to both cavities’ FSR, which therefore need to be matched. Detuning the AOM frequency from the FSR reduces the effective number of passes in the cavity, and therefore increases the SBS threshold. If the top and bottom cavities FSRs are mismatched, then the sensing region becomes longer (the overlap window will “move” in the fiber) and at least one of the cavities will have a mismatched AOM, leading to efficiency reduction. Matching the top and bottom cavities is not difficult, as they both share a common central fiber region where most of the sensing fiber is placed and only need a small independent length to be matched. This adjustment can be made with a tunable optical time delay device (ODL in the lower branch of the sensing oscillator in Fig. 2) and should not need further adjustment during operation since the FSRs are then matched.

The AOMs are used in the gate configuration (zero order used for transmission), similar to Q-switched lasers. However, typically 10 dB extinction ratio is available in such a configuration. The AOMs used here are driven at 55 MHz. When the AOM receive their respective 55 MHz RF signals, the optical gate is closed (power sent to 1st order). This signal, generated by an RF generator is split, amplified and gated by a home-built electronic circuit. This RF gating is controlled by a dual channel signal generator which generates the ~50 kHz gating pulses.

4. Results

By observing the beat note generated by the nth order Stokes waves, S_n,sens with S_n,ref, enables the detection of temperature in different regions along the 4km fiber bundle. Figure 4 shows the variation of the beat frequency for the 2nd Stokes (blue line) and the 4th Stokes waves (red line) as a function of the temperature of the 2.5 km length of fiber placed in an environmental chamber. The temperature in the chamber is controlled at 70.0 °C while another 1.5 km fiber bundle is kept at room temperature. Top inset in Fig. 4 shows the difference in the beat frequencies for a temperature of 22.8 °C and 70.0 °C for S₄ while the bottom curve shows the change for S₂. As can be seen in Fig. 4, sensing with the 4th Stokes wave is more sensitive than using the 2nd Stokes order by a factor of two.

 

Fig. 4 A 2.5 km (area shown by the pale grey rectangle) fiber bundle is kept at 70.0 °C while the rest of the fiber (1.5 km) is maintained at room temperature of 22.8°C. The sensor has a resolution of ~225m (shown by the darker grey rectangle). The temperature sensing signal generated by the 4th Stokes wave is compared with the 2nd Stokes wave shown by the red and blue curves, respectively. The insets show the beat frequencies for the 2nd (bottom inset) and 4th Stokes (top inset), both for a temperature of 22.8 °C (reference oscillator) and 70.0 °C (sensing coil).

Download Full Size | PDF

Figure 4 shows this principle as 2.5 km SMF-28 of fiber is placed in an oven at 70.0 °C while the remaining fiber of the sensing oscillator (1.5 km) is kept at 22.8 °C as is the reference oscillator. This temperature corresponds to room temperature. For the same temperature difference (∆T = 47.2 °C), S₄ is shifted from the reference Brillouin frequency shift by 190 MHz which corresponds to a 4.02 MHz / °C variation. S₂ however, shifts by 94 MHz which is equivalent to 2 MHz / °C corresponding to half that for S₄, but twice the typical temperature sensitivity of Brillouin scattering based sensors of 1.1 MHz / °C using the first Stokes. This means that a smaller variation of temperature could be detected more rapidly, since the frequency shift is larger using the 4th rather than the 2nd order. It should be noted that the 0 °C temperature variation (both reference and sensing oscillator at room temperature) has a beat frequency offset of 2.5 GHz. This is because the reference and probe fibers were judiciously selected to offset the beat frequency away from the DC level, separating the 2nd and 4th Stokes beat note for easier measurement. The reference fiber is a highly nonlinear fiber with small core area from Fibercore with a Brillouin frequency shift of 9.6 GHz while the sensing fiber is a standard single mode fiber (SMF-28 from Corning with a frequency shift of 10.85 GHz. Therefore, the beat frequency for S₂ is centered at 2.5 GHz while the beat note for S₄ is at 5.0 GHz at room temperature as shown in the bottom and top inset of Fig. 4.

Figure 5 shows a distributed sensing measurements performed with the system for different temperature variation detected with S₂. The inset in Fig. 5 shows the displacement of the beat signal between S_{2, sens} and S_2,ref for various temperatures ranging from 22.8 °C to 90 °C. The different edges (rise and fall) position from these different measurements are within the resolution of the system (~225 m).

 

Fig. 5 Distributed sensing measurement at various temperatures ranging from 22.8 °C to 90.0 °C using the 2nd Stokes order. The inset shows the beat signal between S_2,sens and S_2,ref for this temperature range.

Download Full Size | PDF

Aside from the low extinction ratio of the AOMs limiting the spatial resolution and BP power that could be injected, another important limitation that was noted is ASE noise from the erbium amplifiers. Since the upper cavity is seeded with the BP, the signal to noise (SNR) ratio remains good and the Stokes remain dominant above the ASE level. However, for the lower cavity, it is seeded with a lower-power Stokes and the SNR is very poor, a problem which becomes critical when reducing the gate-time as shown in Fig. 6.

 

Fig. 6 OSA spectra of the even (in blue) and odd (in red) Stokes generation. Even Stokes orders are clearly discriminated, while odd Stokes are below the ASE noise under favorable generation conditions.

Download Full Size | PDF

Under unfavorable conditions (short gate time, high Stokes orders), the ASE optical noise level becomes dominant compared to the odd Stokes. Although the spatial discrimination is maintained, the temperature induced frequency shift attributed to the odd Stokes seems to disappear dividing the sensitivity by two. Hence, an unwanted beat note at the ESA spectra appears with half the expected beat note. For instance, when looking at the 2nd Stokes, the additional beat note appears like a normal 1st order Stokes wave which has a 1 MHz / °C variation with temperature instead of 2 MHz / °C for the 2nd Stokes. The blue curve shown in Fig. 7 represents a case when a proper generation of S₂ is used as higher sensitivity distributed temperature sensor, and the dotted line represents the case when SBS is not generated properly within the AOM due to a reduced time-window and a loss of the sensitivity is observed for the same Stokes order.

 

Fig. 7 Demonstration of a case when proper temperature sensing is performed using higher order Stokes wave (blue curve), and when SBS is generated from ASE instead of from a cascaded process of SBS. The inset represents beat frequencies for T = 22.8 °C (in blue), and T = 70.0 °C with proper cSBS generation (in red), and for T = 70.0 °C under poor SNR conditions when ASE is dominant for the odd Stokes wave (in grey).

Download Full Size | PDF

The inset of Fig. 7 depicts three beat-notes in different situations; in blue, when the system detects a constant temperature (same as the reference) which gives the reference beat frequency note. In red, the correct parameters generate a 2nd Stokes wave properly within the AOMs gate time. Finally, in grey, the beat frequency is mostly dominated by SBS generated outside the AOMs temporal gate-times due to the poor choice of parameters (BP, in-cavity EDFA, wrong AOM modulation frequency, polarization, etc). A weak peak (right in grey) is observed, but it is lower than the middle one and sometimes even disappears completely, i.e. no SBS is generated within the AOMs gate window.

In this paper, we were able to achieve an electrical gate of ~600 ns and 1500 ns, which yield pulses of ~100 ns and ~900 ns due to the influence of the mode-locking regime in the cavity. The convolution of such pulses gives a time overlap of ~1100 ns which corresponds to about 225 m in terms of spatial resolution over a total sensing distance slightly over 4 km. Figure 8 shows the pulse width of the AOMs.

 

Fig. 8 Temporal gating time of both AOMs. In red the time window is narrow at around 100 ns, while in red the gating time is longer at approximately 900 ns which leads to a convolution product of 1100 ns leading to a spatial resolution of 225 m.

Download Full Size | PDF

5. Discussion

The novel technique to generate a distributed temperature sensor is promising for high sensitivity measurements, but currently, has some limitations which can be overcome by use of better components. Firstly, since it is a dual-ring cavity, the top fiber length must be adjusted properly to match the bottom one, as mentioned before. This can be done simply by a fine adjustment during initial setup of the system. The sensing fiber also changes slightly in optical length with temperature, altering the common FSR. This change is the same for both cavities as they share the same FUT. Although such change was noted to be very small, even for a large temperature variation of 70°C over ~1-2 km, the slight detuning does reduce the SBS generation efficiency. A slight adjustment to the AOM modulated frequency must therefore be made to maximize the signal during operation. This could be done actively to adapt to any situation by simply maximizing the signal.

Polarization is also an important factor, as it affects the SBS gain [20], therefore adding a polarization controller in each arm of the oscillator helps in the cSBS generation. Although this is not critical, it can be very helpful to optimize SBS generation. AOMs were used partly because they are polarization independent, compared to electro-optic modulators (EOM) which are polarization sensitive. Polarization control would be critical and complex when using EOMs instead.

The resolution of the system is limited by the intra-cavity gain and injected BP power: increasing the former increases ASE noise while increasing the latter allows SBS to be generated in the whole cavity, eliminating spatial discrimination. A potential solution is to increase the tolerable level of BP power which can be injected. Having more BP power would help generate high order Stokes waves within a shorter region, thereby improving spatial resolution. The AOMs’ extinction ratio (attenuation in the OFF-state) of ~10dB was not sufficient to perfectly suppress SBS generation outside of the sensing region when the BP was increased further. To overcome this problem, one would have to use AOMs with a better extinction ratio (for example using two AOMs in series with 1st order followed by a negative 1st order diffraction leading to no net frequency shift from diffraction, but with around 50dB of extinction). This is currently under investigation.

Noise from the ASE and the large number of cavity modes under the bandwidths of the SBS orders affects the performance of the system and thus limits the effective increase in sensitivity. If the noise gets stronger as the Stokes order increases, which seems to be the case comparing S₄ and S₂ from the inset in Fig. 4, it leads to a loss of resolving power of the sensor. However, Liu et al. [15] in their non-distributed Brillouin sensing cavity, overcame this problem by using an un-pumped erbium fiber as a saturable absorber allowing a single longitudinal mode to oscillate within each Stokes bandwidth and demonstrated that the accuracy of the temperature measurement increased from 0.2 °C to 10⁻⁶ °C from S₁ to S₃. Implementation of this modification in our DTS system should lead to a similar improvement as well, and this is currently being investigated.

Also, the actual fall/rise time of the AOM was 70 ns which limits the minimum resolution achievable. Faster AOMs could help decrease the spatial resolution up to around 10 ns (typical AOMs and phonon lifetime in typical optical fiber [20]. Shorter gate times than 10 ns makes it very difficult to generate SBS. Using other encoding techniques such as proposed by Bao et al. [10], could potentially lower the effective spatial resolution tremendously and could therefore, achieve a higher sensitivity with ultra-high resolution.

A more complex electronically controlled AOM gating scheme could potentially allow to use one oscillator as both the reference and sensing oscillator.

Finally, using less noisy Raman gain, the ASE could be reduced, therefore allowing sufficient gain for cSBS generation in the cavity.

6. Conclusion

In this paper, we have demonstrated a technique to combine two concepts: use of high order Brillouin Stokes waves to increase the sensitivity of temperature sensing, and a novel scheme to implement distributed sensing with the high-order SBS Stokes waves. We have demonstrated an increase of 4x over the standard sensitivity achieved in typical Brillouin distributed sensors such as BOTDR and BOTDA systems. The distributed measurement demonstrated had a spatial resolution of approximately 225m (mostly limited by the rise / fall times and extinction ratio of the AOMs used in the experiments) with more than 4 km sensing reach. Temperature detection from 22.8 °C to 90.0 °C was shown using a 2nd order Stokes wave which could easily extend to higher temperatures. A complete distributed sensing scheme using 4th order Stokes wave for a discrete variation of temperature from 22.8 °C to 70.0 °C has also been shown with twice the sensitivity of the 2nd Stokes. The poor extinction ratio of the AOMs and the ASE noise of the amplifiers limited the increase in sensitivity in our current system as generating truly distributed higher order Stokes wave was not possible.

To overcome these problems, dual AOMs with a 1st diffracting order followed by negative 1st diffracting order could improve the extinction ratio to over 50 dB, assuring SBS generation only within the AOMs gate-times. ASE noise from the erbium-doped amplifiers could be solved using Raman amplification with its lower noise figure. This is currently under investigation.

This first prototype of fully distributed high sensitivity temperature sensor is promising and could benefit industrial applications as it would help provide an early warning of subtle unwanted variation in temperature. Fully distributed strain measurement using the same principle is also being investigated in our laboratory.