1. Introduction

Distributed and quasi-distributed fiber-optic temperature sensors have been a topic of exhaustive research for many years. These efforts resulted in a range of commercial applications in various practical fields such as structural monitoring, oil well monitoring, aerospace applications, etc.

Various fully distributed temperature-sensing systems have been successfully demonstrated in the past. Most distributed sensor systems are based on the temperature dependence of Raman [1,2], Brillouin [3–5] and Rayleigh [6,7] scattering. Alternatives based on Sagnac [8,9] effects were also successfully demonstrated in practice. The main drawback of those systems is complex and costly optical signal processing. Therefore quasi-distributed temperature measurements, where a number of discrete sensors are located down the fiber, can present more practical and economical solutions in variety of practical applications. Quasi-distributed systems require less complex and less expensive signal processing relative to the fully distributed versions. Fiber Bragg gratings [10–12] and Fabry-Perot interferometers [13–15] can be, for example, configured in quasi-distributed temperature sensor arrays. However, the complexity of signal processing in those systems is still considerable and cost prohibitive in many practical instances.

This paper presents a design of an all-fiber polarimetric quasi-distributed temperature sensor that utilizes a standard telecommunication optical time domain reflectometer (OTDR) as a signal processing system. Besides commercially available single polarization (SP), high birefringence (HB) fibers and the standard telecommunication OTDR, the proposed system does not require any additional components and therefore presents a potentially cost effective alternative to existing systems.

2. Sensor design

A common (single point) polarimetric sensor [16, 17] utilizes a bulk or fiber input polarizer, a lead-in polarization maintaining (PM) fiber, a section of sensing highly birefringent (HB) fiber spliced at 45° relative to the lead-in fiber, a lead-out polarization maintaining (PM) fiber placed at 45° relative to the sensing fiber and an output bulk or fiber polarizer. This configuration forms a polarimetric interferometer that can be interrogated by conventional interferometric demodulation techniques, such as white light interferometry [18–20].

The appropriate redesign of this configuration can allow for the formation of an all-fiber, compact, and low loss intensity modulated sensor structure. Such a structure can be stacked into a sensor array that can be interrogated by a standard telecommunication OTDR.

A proposed sensor array (Fig. 1) is formed first by a short section of single polarization fiber that is followed by a number of sensor segments that generate temperature dependent loss. The array is concluded by a section of PM fiber with sufficient length allowing the OTDR to read the Rayleigh backscatter signal level. A basic sensor segment of the proposed sensor array is shown in Fig. 2.

 

Fig. 1. Quasi distributed sensor array

Download Full Size | PDF

 

Fig. 2. Structure of the individual sensor segment

Download Full Size | PDF

The sensor segment (Fig. 2) is formed from a lead (transmission) PM fiber that connects individual sensor segments and can be of any practical length (typically 5–1000m) that allows for reading of the Rayleigh backscatter signal by the OTDR. The lead fiber is followed by a short section (typically 1 cm) of sensing HB fiber that is spliced to the lead fiber at an axial rotation angle Φ₁. This excites both polarization modes in the HB fiber that interfere along the sensing fiber. The sensing fiber length is chosen in a way that the phase between the polarization modes does not change for more than π over the target temperature measurement range. This assures unambiguous response of the sensor within the desired measurement range. The sensing fiber is then followed by a short section of the single polarization fiber (typically 1 m) that is spliced to the sensing fiber at an axial rotational angle Φ₂. As described later in detail, angles Φ₁ and Φ₂ determine the interference modulation depth (fringe visibility) that needs to be well controlled in a loss interrogated quasi-distributed sensor system network. The single polarization fiber is then spliced to the lead fiber of the next sensor segment and the axes of both fibers are aligned. The single polarization fiber therefore acts as an analyzer for the first sensor segment and polarizer for the next sensor segment. An exception is the first sensor where the polarization of light is assured by a single polarization fiber placed between the OTDR and the first sensor segment.

3. Sensing fiber length

The phase change, Δϕ among polarization modes in stress induced HB fiber due to the temperature change ΔT can be expressed as [21]:

where k is the constant that is determined by the fiber profile design, fiber birefringence and wavelength. For bow-tie fibers this value is in the range of a few (typically 0.5–5) rad.K^-1.m^-1. To provide maximum phase change of π over the target sensor temperature range, the length of the sensor fiber is chosen according to:

For a 100 °C temperature range and typical commercial bow-tie fiber (k=1 rad.K^-1.m^-1 at 1310 nm) the sensor fiber length is about L_sens= 1 cm. 1 cm also proved to be the minimal practical sensor fiber length. Attempts to reduce the sensor length significantly below 1 cm led to an increase in noise that is most likely attributed to the cladding or leaky modes that do not radiate away in very short segments of fiber (similar effects were reported when single mode fiber were used as modal filters [22]). Choosing sensing fibers with lower temperature sensitivity or simply fibers with lower birefringence can provide broader, unambiguous temperature ranges. The selection of L_sens also determines sensor sensitivity Δϕ/ΔT=k L_sens.

While the selection and rough adjustment of the sensing fiber length L_sens in accordance with eq. (2) provides the desired unambiguous temperature measurement range and sensitivity, fine adjustment of the L_sens tunes the sensor operating point and thereby provides an unambiguous and quasi-linear relationship between sensor loss and the measured temperature within the target temperature range. In other words, L_sens needs to be fine adjusted in a way to place the sensor in a quadrature point where the sensor temperature corresponds to one half of the maximal temperature. Otherwise, the interferometer cosine transfer function goes through its maximum (or minimum) somewhere in the middle of the target temperature measurement range and the loss-temperature characteristics become ambiguous. Fine-tuning can also assure all sensors in the network will have the same response characteristics and that no further calibration is need. The fine adjustment process is described in the experimental section.

4. Adjustment of Φ₁ and Φ₂

Adjustment of Φ₁ and Φ₂ allows for control over maximum temperature induced loss that occurs at the individual sensor segments. Good control over this loss is essential for the optimum design of a quasi-distributed sensor array. Selection of larger modulation depths results in larger changes of loss at an individual sensor segment that can be therefore interrogated with higher resolution. Unfortunately, larger modulation depths also increase maximum cumulative loss in the array and therefore limit the number of sensor segments that can be configured into the network. Adjustment of Φ₁ and Φ₂ can therefore be used to achieve the optimum compromise between desired number of sensors in the array and sensor resolution. Furthermore, different pairs of Φ₁ and Φ₂ can be set at the beginning and at the end of an array to program the maximum losses of individual sensors within the array for the optimum network performance (for example, deeper modulation of loss might be desired towards the end of the network to achieve the same temperature resolution as at the beginning of the network due to the weaker optical signal at the end of the array).

By applying the Jones formalism for proposed polarimetric sensor configuration, and assuming that only one polarization mode is excited in the lead sensor segment fiber, the light intensity at the output of the proposed sensor segment can be expressed as:

Since we are interested in modulation depth that occurs during a full temperature cycle, we need to consider extremes of the function (3), e.g. when Cos(ϕ)=±1. Extreme values of I_out out can be then expressed as

There are various Φ₁ and Φ₂ combinations that can provide the desired sensor modulation depth. However, to minimize the sensor loss, only the combinations of Φ₁ and Φ₂ that provide maximum transmission at the target modulating depth should be used.

The global maximum of function (4) (I_extreme=I₀) occurs when:

By discarding the solutions for Φ>π, two expressions for output intensity can be written. For Φ = Φ ₁ = Φ ₂:

and for Φ = Φ ₁ = -Φ ₂

The loss modulation depth mod, e.g. the ratio between the minimum (cos(ϕ)=1 for (6) and cos(ϕ)=-1 for (7)) and maximum (cos(ϕ)=-1 for (6) cos(ϕ)=1 for (7)) intensity, is the same for both cases and can be expressed from (6) or (7) as:

Finally, the required fiber axial rotation angles Φ=Φ ₁=±Φ ₂ expressed as a function of the desired loss modulation depth mod are:

5. Experimental results

A practical demonstrator having three temperature sensors was built and tested with a conventional telecom OTDR at 1310 nm. We chose to build sensors for temperatures ranging from 0 to 100 C with a target modulation depth of 0.9 dB. Also, we fine-tuned the sensor fiber length in a way to achieve a minimum loss at 0 C and a maximum loss at 100 C.

Sensors were made with Corning’s commercially available SP fiber [23,24] and Fibercore Inc. Bow-Tie PM fiber (HB 1250) both designed for 1310 nm. HB1250 had a temperature sensitivity of k=1 rad.K-¹.m^-1 at 1310 nm. It follows from (2) that L_sens needs to be 1 cm to achieve an unambiguous response in the 0-100 C temperature range.

The Corning single polarization fiber suppresses undesired polarization mode below -40 dB while the attenuation of the supported polarization modes is less than 0.1 dB. Both values apply for a 1 m section of the fiber. The extinction ratio is therefore more than sufficient to provide high-resolution operation of the polarimetric system [17]. The main drawback of the single polarization fiber is in its small and highly elliptical modal field, which limits the minimum loss that can be achieved during fiber splicing. The splice loss can be improved by thermal expansion of the core. We tested splice losses with different types of PM fibers. Each fiber required optimization of the fusion parameters. For the example where we spliced a bow-tie HB1250 to single polarization fiber, we used an Ericsson FSU 975 PM-A splicer. The following fusion parameters were used: Fusion Time 1 = 0.3s, Fusion Current 1 = 10.5mA, Fusion Time 2 = 3.5s, Fusion Current 2 = 16mA and Fusion Time 3 = 0s. Fusion time 2 is long, and provides thermal expansion of the core. This time needs to be longer if the mode field diameter of the PM fiber is larger and vice versa (Corning recommends 20s for fusion time 2 when SP fiber is spliced to standard single mode fiber). After proper fusion splice parameter optimization, the SP to PM fiber splice loss proved to be between 0.3 and 0.35dB for all tested PM fibers. The achieved minimum loss of the entire single sensor segment that included a total of three splices (two splices between SP and PM fibers and one PM to PM splice) and loss of the supported mode in SP fibers was around 0.9 dB. This loss was achieved by off the shelf fibers and could be most likely further reduced if either single polarization or HB fiber would be optimized to better match each others modal fields.

The configuration Φ=Φ₁=-Φ₂ was chosen and it followed from (9) that Φ₁ and -Φ₂ need to be set to 12.8 degrees in order to obtain the desired modulating depth of 0.9 dB. The rotational alignment of SP to PM fiber splices was performed by an Ericsson FSU 975 PM-A fusion splicer in manual mode with the help of a polarization analyzer. This is because the current PM fusion splicers do not support splicing of single polarization fibers. However it can be expected that the new single polarization fiber data will be included in the new PM splicer’s database and will provide fast and accurate automatic alignment as is the case for standard PM fibers.

 

Fig. 3. Typical transmission temperature characteristics of the sensor segment after successful splicing

Download Full Size | PDF

Figure 3 shows an example of the single sensor segment transmission versus temperature after successful completion of the rotational alignment and splicing procedure. The modulation depth of the sensor was 0.91 dB. Fine tuning of the sensor length was performed to set the minimum loss at 0 C and maximum loss at 100 C. The precision tuning was accomplished by heating and stretching of the sensing fiber length. The heated region was sufficiently wide (at least 5 mm) so that the formation of the slightly tapered region did not cause mode coupling within sensing PM fiber. This was achieved by application of an oscillating flame torch [25,26] (the sensing fiber was heated over an oscillating flame torch and pulled apart by a linear micro stage). The tuning process was also accomplished in a few iterative steps to avoid overstretching of the sensing fiber. Before tuning, the temperature characteristic was first measured and the sensing fiber length correction was calculated according to:

where B_L is fiber beat length and Δτ is the desired sensor phase correction.

The fiber was then heated and stretched to a length that was 20–30% shorter than the predicted correction length (Equation 11) to prevent sensor overstretching. The temperature characteristic was re-measured and a new sensor length correction was calculated. The process was iterated until a desired sensor characteristic was achieved. Typically we need two to three iterations to reach the desired characteristics. Figure 4 shows an example of successful tuning of the sensor in two iterative steps. We used hand driven linear micro stages to perform fiber stretching. However, automated and motorized micromanipulation could probably be used to achieve tuning in a single step.

 

Fig. 4. Tuning of the sensor

Download Full Size | PDF

Figure 5 shows a network of three sensor segments interrogated by an OTDR. The lead HB fiber between the sensor was about 15 m long. As an example, the temperature was kept constant at the middle and varied on the first and last sensor over a 0–100 C range.

 

Fig. 5. (Movie 1,56MB) OTDR response of three sensors at different temperatures.

Download Full Size | PDF

When the (quasi) liner response characteristic of the sensor is assumed, the absolute temperature resolution of the sensor system can be estimated as:

In case of our practical demonstrator, the sensor resolution should therefore correspond to about 111 C/dB*(OTDR resolution). The experimentally determined resolution (measured at 50 C) was better than 1 C. In the experiment we used an older type of OTDR (GNN Nettest model TD-385 with module TD-362) that has declared loss resolution of 0.01dB. Modern OTDRs have abut 10 times better resolution and it can be therefore expected that the system would detect proportionally smaller changes of temperature if contemporary version of OTDR would be employed.

The achievable sensitivity for the proposed scheme (assuming liner characteristics) can be described as:

In particular demonstrator case the estimated sensitivity corresponded to 0.009 dB/C. The sensitivity can be increased by reducing unambiguous temperature range (e.g. using longer sensing fiber) or increasing sensor modulation depth by increasing Φ (that will increase overall loss in the network and thereby limit the number of sensors in the network). Sensitivity and resolution are therefore tradeoff between the unambiguous temperature range and maximum affordable loss per senor segment (e.g. number of sensors that can be addressed by the OTDR).

Although the proposed scheme might appear difficult in practical preparing, there are various possibilities for improvements. The PM to SP fiber alignment during splicing, which was done manually, could be easily replaced by full automatic procedure (appropriate data needs to be included in splicer’s rotational alignment algorithm which is standard task/requirement for PM splice manufacturers, especially since SP fiber provides very characteristics and well identifiable side view pattern when rotated). The fine tuning alignment based on heating and stretching could be also replaced by precision cleaving. If we again assume liner sensor response the required accuracy of the sensor segment length can be estimated as:

where 1/2 of the beat length presents the physical variation in sensor fiber length that brings the sensor through the full measurement range (e.g. from min to max loss), maximum initial offset is discrepancy between the target loss and desired loss at given (production) temperature.

For example, if we assume 100 C temperature range, fiber with the beat length of 1 mm and we need to assure that all produced sensor characteristic is within 1 C tolerance, the sensor fiber should be cleaved within (1/2 * 1000 μm * 1C /100C =) 5 um tolerance range, a value that is achievable by customized precision cleaver construction. Finally, there is a need to align the polarization of the OTDR source to the axis of the input SP fiber. While utilization of a polarization controller is an option, more economic and practical solution is preferred. In reality the full polarization alignment is not required, what is needed is only reasonable stable polarization state of the light at the input SP fiber in order to prevent significant loss fluctuations (fading) at the connection between and OTDR and SP fiber (OTDR usually contains laser diode that is pigtailed with standard single mode fiber and therefore the output polarization state can randomly fluctuate over time). There are various solutions to this problem. For example using a PM fiber pigtailed laser diode instead of standard single mode fiber requires a minimum customization of a standard OTDR. Even more economic solution could be achieved by introduction of depolarizer between OTDR and input SP fiber. For this purpose a cheap all-fiber Lyot-type depolarizer can be build (the coherence length of pulsed laser diode in telecomm OTDR is usually quite short, typical at the order of mm or less, meaning that only few meters of PM fiber is needed to build effective depolarizer). The later solution will add 3 dB of loss which is certainly acceptable in this application.

6. Conclusions

This paper presented a design for a quasi-distributed polarimetric temperature sensor network. The sensor network was entirely built out of commercially available SP and HB fibers and was interrogated by standard telecommunication OTDR. Since the concept does not require any other optical or optoelectronics components, the presented solution presents a potentially cost effective alternative to existing concepts.

An all-fiber polarimetric interferometer setup and the temperature dependence of birefringence in HB fiber were used to create sensing segments that exhibit temperature dependent loss. The modulation depth of the loss and an unambiguous temperature range were programmed through proper rotational alignment of SP and HB fibers and through coarse adjustment of the sensing HB fiber length. The sensor unambiguous temperature operating range and high production repeatability were achieved through the fine tuning of the sensing HB fiber length. This allows for optimization of the sensor network to specific practical application requirements. The (undesired) insertion losses resulting from splice losses and losses in SP fiber averaged around 0.9 dB, and could be further improved by additional fusion splice parameter optimization and/or tailoring of the HB and SP fiber profiles to better match each others modal fields.

The design was demonstrated on a sensor network, built with sensors that had a temperature range 0–100 C and generated a temperature dependent loss of 0.9 dB per sensor segment. The maximum total loss (including splice losses and modulation loss) per sensor was therefore 1.8 dB. Standard telecommunication OTDR with typical 40 dB dynamic range and 0.001 dB resolution could therefore address more than 20 such sensors with a resolution up to 0.1 C.