1. INTRODUCTION

Tunable diode-laser-absorption spectroscopy sensors can provide nonintrusive, in situ, species-specific measurements of gas properties in a wide range of practical environments (e.g., engines, power plants) [1–6]. During operation, the wavelength-specific transmission of laser light through an absorbing gas is measured and the absorption is compared with spectroscopic models to infer gas properties. In most applications, laser light is pitched across a line of sight with a laser or pitch fiber (i.e., the transmitter) on one end and a photodetector (i.e., receiver) on the other end. While this approach enables high-optical throughputs and, thus, high signal-to-noise ratio (SNR), it is not applicable to many systems with limited optical access and prohibits stand-off gas monitoring.

To address this issue, several researchers have developed absorption spectroscopy sensors that operate by detecting laser light that is backscattered off native surfaces [7–9] or surfaces embedded into probes [3,10–12]. Dubinsky et al. [7] first showed that reliable frequency-modulation spectroscopy (FMS) spectra could be obtained when collecting backscattered laser light and demonstrated this with measurements of iodine vapor. Later, Wainner et al. [8] developed a battery-powered, handheld FMS sensor for stand-off detection of methane at up to 30 m and with a sensitivity up to $4\text{ppm-m}/\sqrt{H}z$. Most recently, Wang and Sanders [9] demonstrated the use of an off-axis (i.e., pitch and catch are not coaxial) wavelength-modulation spectroscopy (WMS) sensor for detecting ${\mathrm{H}}_2\mathrm{O}$ near 1350 nm and found that $\mathrm{WMS}\text{-}2f/1f$ spectra with high SNR ($\approx 400$) could be obtained off rough surfaces with low light collection [$\approx 500$ parts per million (ppm)].

Unique to our sensor is: (1) the use of a handheld, fiber-coupled transmitter and receiver that is (2) packaged in a coaxial fiber bundle with a single detached lens, and (3) the use of two-color scanned-$\mathrm{WMS}\text{-}2f/1f$ to enable (4) calibration-free stand-off measurements of gas temperature, pressure, and composition at (5) 200 to 2000 times greater bandwidth compared to past work. In our opinion, the primary advantage of our design, compared to past work, is that use of a fiber-coupled transmitter and receiver enables the laser, detector, and associated electronics to be detached, thereby reducing the weight and size of the light delivery and collection units, which (1) enables more practical utilization as a hand-held or wearable sensor, and (2) could facilitate the integration of such a sensor into a real-world system with limited optical access (e.g., an engine). Here we present the design and initial demonstration of this sensor with measurements of temperature, pressure, and composition in a flame at a stand-off distance of 15 cm and measurements of ${\mathrm{CH}}_4$ at a stand-off distance of 10 m. By using scanned-$\mathrm{WMS}\text{-}2f/1f$ with fast-wavelength modulation ($\approx 100\mathrm{kHz}$)) we show that the influence of cavity noise (inherent to our coaxial pitch–catch arrangement) is reduced to acceptable levels, enabling measurements of absorption transition integrated areas and collisional widths with uncertainties of 1.5% and 3%, respectively, in both turbulent and quiescent gases.

2. SCANNED-$\mathsf{WMS}\mathsf{\text{-}}\mathsf{2}\mathsf{f}/\mathsf{1}\mathsf{f}$

In scanned-$\mathrm{WMS}\text{-}2f/1f$, the nominal wavelength of a laser is injection-current scanned across an absorption transition while the instantaneous wavelength is rapidly (compared to the scan) sinusoidally modulated. $\mathrm{WMS}\text{-}2f/1f$ spectra are extracted from the detector signal during postprocessing using a digital lock-in filter [13]. By using $1f$-normalization, the WMS spectra are independent of the detector gain and DC light intensity reaching the detector [14]. This attribute is vital to the operation of this sensor since the backscattered light intensity reaching the detector can vary widely during operation.

The scanned-$\mathrm{WMS}\text{-}2f/1f$ spectral fitting routine developed by Goldenstein et al. [15] (built off the simulation strategy of Sun et al. [16]) was used to provide calibration-free measurements of temperature, pressure, and composition. In this method, the time-varying transmitted light intensity, $I_t(t)$, impinging on the detector is simulated according to Eq. (1),

3. SENSOR DESIGN

A. Hardware

Schematics of the fiber bundle (Neptec OS) and experimental setups are shown in Fig. 1. The bundle consists of one single-mode fiber (Corning SMF-28) for transmitting laser light and six multimode fibers (105 μm, $\mathrm{NA}=0.22$) equally spaced around the perimeter for receiving the backscattered laser light. These parameters were dictated by the availability of the fiber bundle and, as a result, were not selected as an optimized design. On the transmitting/receiving end, all fibers are enclosed in a single FC-PC termination. On the opposite end, the six multimode fibers are split off into a single arm that directs the collected laser light to an InGaAs detector (Thorlabs PDA10CS) and the SMF-28 fiber is connected to the diode laser(s) directly or via a fiber combiner. The fiber bundle was mounted inside a lens tube with a single anti-reflection-coated plano–convex lens ($d=25.4\mathrm{mm}$, $f=60\mathrm{mm}$ or $d=50.8\mathrm{mm}$, $f=125\mathrm{mm}$) for a lightweight ($\approx 0.15\mathrm{kg}$), handheld transmitter/receiver. No laser light was detected in the absence of external (i.e., outside the lens tube) backscattering media. For a given stand-off distance, the fiber bundle was positioned at a distance outside the focal length of the lens to focus the laser beam onto the target (i.e., reflector). This was done to maximize the collection efficiency at a given stand-off distance. This adjustment was performed manually by visually monitoring the spot size of a multiplexed visible beam. Figure 2 shows the fraction of light collected as a function of stand-off distance for matte aluminum and paper reflectors. The fraction of photons collected was determined using the known wavelength and measured values of incident (i.e., exiting the pitch fiber) and collected (i.e., exiting the catch fibers) optical powers (measured with a commercially available fiber-coupled power meter). Collection efficiencies were largest using diffuse reflectors (e.g., matte metal, paper, wood, drywall) and smallest for specular reflectors (e.g., polished metal). For both reflectors, the fraction collected decreases with increasing stand-off distance with a slope near 2 (on a log-log scale), consistent with a $1/L^2$ dependence. However, some nonlinear trends were observed, perhaps due to the spot size of the focused beam (1–10 mm in diameter) also varying with stand-off distance. Interestingly, the 50.8 mm diameter lens did not systematically outperform the 25.4 mm lens, but this is likely due to the fact that the longer focal length 50.8 mm lens produced a considerably ($3–4\times$) larger spot size on the target.

 

Fig. 1. Experimental setup for backscattered $\mathrm{WMS}\text{-}2f/1f$ measurements of temperature, pressure, and ${\mathrm{H}}_2\mathrm{O}$ in a flame (left) and ${\mathrm{CH}}_4$ concentration (right).

Download Full Size | PDF

 

Fig. 2. Fraction of incident photons collected as a function of stand-off distance.

Download Full Size | PDF

Distributed-feedback diode lasers (NEL America) producing $\approx 20\mathrm{mW}$ near 1391.7, 1343.3, and 1651.0 nm were used to measure temperature, ${\mathrm{H}}_2\mathrm{O}$, and ${\mathrm{CH}}_4$. Tables 1 and 2 present the most relevant spectroscopic parameters of the transitions of interest. During temperature and ${\mathrm{H}}_2\mathrm{O}$ measurements, the lasers near 1391.7 and 1343.3 nm were injection-current scanned across their respective ${\mathrm{H}}_2\mathrm{O}$ transitions with a 1 kHz sine wave (yielding measurements at 2 kHz due to the up-scan and down-scan) and were frequency-multiplexed with modulation frequencies ($f_m$) of 160 and 200 kHz and modulation depths ($a_m$) of 0.09 and $0.06{\mathrm{cm}}^{-1}$, respectively. Approximately 20 μW (1000 ppm after coupling losses) per laser was collected and the detector gain was set to 30 dB (775 kHz bandwidth). During ${\mathrm{CH}}_4$ measurements, the laser near 1651 nm was scanned across the four blended ${\mathrm{CH}}_4$ transitions near $6057.09{\mathrm{cm}}^{-1}$ [17] with a 100 Hz sine wave and modulated at 80 kHz with $a_m=0.145{\mathrm{cm}}^{-1}$. The detector gain was set to 40 dB (320 kHz bandwidth) and $\approx 100\mathrm{nW}$ (5 ppm) was collected. During all experiments, a 16 bit data acquisition system (National Instruments) was used to generate and collect all signals at a sampling rate of 2 MHz. 0.6 and 20 kHz Butterworth filters were used to extract $\mathrm{WMS}\text{-}2f/1f$ spectra for ${\mathrm{CH}}_4$ and ${\mathrm{H}}_2\mathrm{O}$ transitions, respectively. All modulation depths were chosen to maximize $2f$ signals.

Table 1. Frequency, Strength, and Lower-State Energy of the Absorption Transitions Used^a

View Table | View all tables in this article

Table 2. Collisional-Broadening Parameters of the ${\mathsf{H}}_{\mathsf{2}}\mathsf{O}$ Transition Near $\mathsf{7185.59}{\mathsf{cm}}^{-\mathsf{1}}$^a

View Table | View all tables in this article

B. Selection of Modulation Frequency

In this sensor the transmitted and reflected light rays overlap each other, which can lead to cavity noise analogous to that seen in cavity-enhanced absorption experiments [20]. Figure 3 (left) shows the raw detector signal using the experimental setup shown in Fig. 1 (left) with a scan frequency of $f_s=100\mathrm{Hz}$ and $f_m=10\mathrm{kHz}$ or 160 kHz and equal modulation depths ($a_m=0.09{\mathrm{cm}}^{-1}$). For $f_m=10\mathrm{kHz}$, large amplitude noise indicative of cavity-like resonances is obvious throughout the modulation period and reaches a maximum near the top and bottom of the modulation envelope [see arrows in Fig. 3 (left)] due to $d\nu /dt$ approaching zero. Figure 3 (right) shows that this cavity noise propagates into the $\mathrm{WMS}\text{-}2f/1f$ spectra obtained using a 3 kHz digital lock-in filter, leading to highly distorted $\mathrm{WMS}\text{-}2f/1f$ spectra. In contrast, with $f_m=160\mathrm{kHz}$, the cavity-like noise is heavily reduced due to the $16\times$ larger $d\nu /dt$ and is now only obvious near the top and bottom of the modulation envelope. Similarly, the cavity noise is nearly unobservable in the $\mathrm{WMS}\text{-}2f/1f$ spectra with $f_m=160\mathrm{kHz}$. This reduction in cavity noise via fast wavelength scanning has been noted by others in off-axis-ICOS experiments [20]. Undistorted $\mathrm{WMS}\text{-}2f/1f$ spectra were observed with $f_m>\approx 100\mathrm{kHz}$, complementing the recent findings of Wang and Sanders [9] who showed that WMS with $f_m>100\mathrm{kHz}$ substantially reduces speckle noise. It should be noted that the observed cavity noise can be reduced in the $2f/1f$ signal by using narrower lock-in filters. However, this is not always a viable option for rejecting noise since using narrower filters reduces the sensor’s bandwidth (in the case of fixed WMS) and relatively wide filters (e.g., 20 kHz passband for the 1 kHz scanning here) are needed to pass entire scanned-WMS signals [6]. As a result, the preferred method to reject this cavity noise is fast wavelength modulation.

 

Fig. 3. Raw detector signal for backscattered $\mathrm{WMS}\text{-}2f/1f$ experiment at nonresonant wavelengths (left) and $\mathrm{WMS}\text{-}2f/1f$ absorption spectra (right) with $f_m=10$ and 160 kHz.

Download Full Size | PDF

4. SHORT-PATH DEMONSTRATION: $\mathsf{T}$, $\mathsf{P}$, and ${\mathsf{H}}_{\mathsf{2}}\mathsf{O}$ SENSING

Measurements of temperature, pressure, and ${\mathrm{H}}_2\mathrm{O}$ were acquired in an unsteady propane flame (see Fig. 4) to demonstrate the potential for high-bandwidth single-ended monitoring of practical combustion systems. A rigorous discussion regarding the influence of line-of-sight nonuniformities and design considerations for mitigating their impact can be found in [21]. With the flame on, recorded emission levels reached 0.3 V (10% of the total signal) and varied at frequencies $\ll f_m$. High-fidelity $\mathrm{WMS}\text{-}2f/1f$ spectra were acquired for both transitions throughout the flame with SNRs typically near 200 enabling best-fit $\mathrm{WMS}\text{-}2f/1f$ spectra within 2.5% of the peak signal near the linecenter. The 95% confidence interval (calculated using Matlab’s nlinfit function) of the best-fit integrated areas and collisional widths (used to calculate gas properties) were less than 1.5% and 3%, respectively. For calculating the ${\mathrm{H}}_2\mathrm{O}$ mole fraction in the flame, the path length was estimated to be 2.5 cm (single pass) from the span of visible flame emission. The ambient ${\mathrm{H}}_2\mathrm{O}$ absorption was accounted for in the scanned-WMS model as described in [15]. The transmitted intensity of each scanned and modulated laser was recorded with only ambient absorption present (i.e., the flame off) and was then defined as the laser-specific $I_o(t)$ in the scanned-WMS model used to simulate scanned-WMS signals acquired in the flame. In this case, this method will incorrectly remove the contribution of ambient absorption, across the 2.5 cm portion of the beam path, from the integrated area measured once the flame is ignited. As a result, this portion of the integrated area (calculated using known ambient ${\mathrm{H}}_2\mathrm{O}$ and temperature) was added back to integrated areas measured with the flame on before calculating gas properties.

 

Fig. 4. Example measured and best-fit $\mathrm{WMS}\text{-}2f/1f$ spectra for ${\mathrm{H}}_2\mathrm{O}$ transitions (corresponding to $t=0.1\mathrm{s}$) (left) and time-resolved temperature and ${\mathrm{H}}_2\mathrm{O}$ mole fraction acquired in a turbulent propane flame using the experimental setup shown in Fig. 1 (left).

Download Full Size | PDF

The average steady-state flame temperature was 1470 K, which agrees relatively well with that of uncorrected thermocouple measurements, 1250–1540 K, acquired along the measurement path. The high-bandwidth WMS temperature measurements indicate that the flame temperature oscillates $\approx \pm 175\mathrm{K}$ at near 10 Hz and $\approx \pm 35\mathrm{K}$ at higher frequencies; some of the latter may be attributed to sensor noise. The ${\mathrm{H}}_2\mathrm{O}$ mole fraction oscillates in phase with temperature (see Fig. 4 zoom inset) indicating that these fluctuations result from variations in local combustion efficiency. For comparison, adiabatic flame temperatures for propane-air flames were calculated for equivalence ratios of 0.45 to 0.70 (chosen to produce the measured span of ${\chi }_{{\mathrm{H}}_2\mathrm{O}}\approx 0.07$ to 0.11) using constant enthalpy-pressure equilibrium calculations. The corresponding change in adiabatic flame temperature between these two cases was 472 K, which is similar, given the many embedded approximations, to the corresponding measured temperature oscillation of $\approx 350\mathrm{K}$ (i.e., $\approx \pm 175\mathrm{K}$). It is worth noting that the temperature sensitivity, defined as the unit change in two-color ratio per unit change in temperature, for this sensor varies from 3.65 to 0.65 as the temperature is increased from 298 to 1700 K. If increased temperature sensitivity is desired (e.g., for cases with lower SNR than here), the transition near $7185.59{\mathrm{cm}}^{-1}$ could be paired with that near $6806.03{\mathrm{cm}}^{-1}$ ($E^{\prime \prime }=3291{\mathrm{cm}}^{-1}$) for larger temperature sensitivity, as done in [22].

With the torch at quasi-steady state (0 to 0.75 s), the best-fit collisional width of the transition near 1391.7 nm combined with empirical models for temperature-dependent ${\mathrm{H}}_2\mathrm{O}$-, ${\mathrm{CO}}_2$-, and ${\mathrm{N}}_2$-broadening coefficients (see Table 2) recovered the known pressure (1 atm) to within 1.8% with a $2\sigma$ precision of $\pm 7\%$. We assumed ${\chi }_{{\mathrm{CO}}_2}=0.75{\chi }_{{\mathrm{H}}_2\mathrm{O}}$ when calculating the collisional-broadening coefficient of the mixture.

Near 0.75 s, the flame was turned off to produce a known transient and near 1.2 s the flame briefly reignited. With the flame off, the sensor recovered the ambient temperature, pressure, and ${\mathrm{H}}_2\mathrm{O}$ concentration within $3(\pm 1)\mathrm{K}$, $1(\pm 0.75)\%$, and $1.5(\pm 0.3)\%$, respectively (the known concentration of ambient ${\mathrm{H}}_2\mathrm{O}$ was measured using direct absorption spectroscopy). The larger scatter associated with the pressure measurement in the flame results from the combined uncertainty in temperature, composition, and collisional-broadening coefficients.

5. LONG-PATH DEMONSTRATION: ${\mathsf{CH}}_{\mathsf{4}}$ SENSING

Measurements of methane in the ambient at a stand-off distance of 10 m were acquired to demonstrate the utility of this sensor at long distances. Figure 5 shows time-resolved methane concentration during the formation of a simulated ${\mathrm{CH}}_4$ leak. Initially, measurements were conducted with a 1 Hz bandwidth (100 raw scans were averaged before lock-in filtering) and a detection limit of $5\text{ppm-m}/\sqrt{H}z$ was obtained. The sensor bandwidth was increased to 100 Hz (i.e., no scan averaging) 4 s into the test to resolve the evolution of a ${\mathrm{CH}}_4$ plume formed by opening a balloon filled with methane. After opening the balloon, path-averaged concentrations of ${\mathrm{CH}}_4$ as high as $2\times {10}^3\mathrm{ppm}$ (i.e., $2\times {10}^4\text{ppm-m}$) were measured. The SNR of the $\mathrm{WMS}\text{-}2f/1f$ signal near the linecenter ranged from 21 (with 100 ppm-m of ${\mathrm{CH}}_4$) to 1500 (with $2\times {10}^4\text{ppm-m}$); the noise level was taken from the $\mathrm{WMS}\text{-}2f/1f$ signal at nonresonant wavelengths. The ${\mathrm{CH}}_4$ concentration was calculated using the line strengths of the four ${\mathrm{CH}}_4$ lines given in the HITRAN2012 database [17] (see Table 1).

 

Fig. 5. Time-resolved ${\mathrm{CH}}_4$ concentration during a simulated ${\mathrm{CH}}_4$ leak.

Download Full Size | PDF

6. CONCLUSIONS

A single-ended, fiber-coupled laser-absorption sensor with a portable, handheld transmitter/receiver has been presented. This sensor operates by collecting wavelength-modulated laser light that has been backscattered off native surfaces located 10 cm to 10 m away while collecting 10,000 to 1 ppm of the incident photons, respectively. Calibration-free measurements of temperature, pressure, and composition in an unsteady propane flame and time-resolved ${\mathrm{CH}}_4$ monitoring were presented. By using scanned-$\mathrm{WMS}\text{-}2f/1f$ spectral fitting with modulation frequencies near 100 kHz and above, the influence of cavity-like noise was mitigated, enabling the determination of integrated absorbance areas and collisional widths with uncertainties of 1.5% and 3%, respectively.