Urban open-air chemical sensing using a mobile quantum cascade laser dual-comb spectrometer

Urban trace-gas identification and quantification is a challenging task. Few other environments exhibit such a dynamic and unpredictable background gas matrix with spatial and temporal variations that evolve with human activity not only on a diurnal cycle but also on hourly, weekly, monthly and annual cycles.¹ In addition, the urban landscape itself adds challenges with buildings, structures, and vehicles that cause turbulence, constrain air flow, alter the wind direction, and restrict the line-of-sight. However, since more than half of the world’s population resides in densely populated areas,² urban atmospheric monitoring is increasingly important. Urban trace-gas measurements not only assess the general air quality but also provide valuable information for public health and safety, e.g., detection of unauthorized emissions, hazardous gas leaks, and clandestine chemical production of drugs or explosives. Due to the low abundances, short lifetimes, and spatially varying levels of most urban trace-gases, sensors designed for this environment need to have high sensitivity and good spatial and temporal resolution while covering large areas in short periods of time. Due to the complexity of urban trace-gas sensing, no individual sensor type is able to fulfill all requirements, which is why a combination of static and mobile point sensors,^3,4 remote detection systems,^5–7 satellites,⁸ and airborne sensors⁹ is used. Laser-based systems play a critical role in this process since they possess many of the desired qualities for urban sensors: they can be configured both as point sensors and remote sensors; they achieve good sensitivity and selectivity; they provide high temporal resolution, and they operate with very low maintenance requirements. Many different types of laser-based sensors have been employed for trace-gas monitoring to date, including both selective tunable laser sensors^10–13 for single/few species detection and broadband sensors for multispecies assessments.^4,6,14–16 In this work, we demonstrate a broadband mid-infrared sensor based on dual-comb spectroscopy (DCS)^17–21 using compact and all-electrically controlled semiconductor frequency combs.²² This spectroscopic technique is in many ways ideal for urban monitoring applications since it provides multispecies detection with good sensitivities and fast acquisition times, all in an optical setup without moving parts. In the near-infrared, dual-comb spectrometers based on fiber frequency combs have been successfully deployed for field-campaigns measuring emissions from, e.g., oil and gas production²³ and agricultural gas flux.²⁴ Recently, field-deployed mid-infrared dual-comb spectrometers have also been demonstrated,^25,26 targeting water isotopes²⁷ and greenhouse gases.²⁸ Although few dual-comb spectrometers have been field-deployed to date, the increasing availability of commercial turn-key mode-locked frequency comb systems has made this technology more widely accessible, opening the possibility of dual-comb spectrometers for applications in both field and laboratory settings.

For urban field applications, a large subset of the targeted gases are low abundance polyatomic molecules, most of which lack (or have very weak) absorption features in the near-infrared part of the electromagnetic spectrum. In order to optically detect these species, access to the mid-infrared wavelength range is crucial. A few different approaches exist to realize mid-infrared optical frequency combs, such as non-linear frequency conversion by difference frequency generation (DFG),^29,30 optical parametric oscillators (OPOs),^31,32 and microresonators,^33,34 all of which have demonstrated chemical sensing capabilities. An alternative is to use the quantum cascade laser (QCL)²² or interband cascade laser (ICL) frequency comb platforms,³⁵ which are chip-scale, electrically pumped semiconductor lasers that inherently generate coherent mid-infrared comb radiation, albeit at the cost of large mode spacing (5–10 GHz) and a narrow spectral coverage (<100 cm⁻¹). If the application allows for such trade-offs, a compact system based on QCLs or ICLs is potentially an attractive solution.

In this work, we use two QCL frequency combs to demonstrate field-deployment of a mid-infrared dual-comb spectrometer operating in a retroreflector based open-path remote sensing configuration.^17,18 The QCLs cover 970 to 1010 cm⁻¹ (9.9–10.3 μm) with a mode spacing of 0.328 cm⁻¹ (9.83 GHz). The system was mounted in a vehicle to allow daily relocation and was transported to the downtown Boston (Massachusetts) area, where system stability assessments and measurements were conducted over a five day period. Remote sensing was accomplished by targeting retro-reflectors (63.5 mm clear aperture) up to distances of 230 m from the sensor. The retro-reflectors were aimed at using a computer-controlled motorized gimbal together with an imaging system that allowed for semi-autonomous operation with pre-determined positions of the targets. To test the instrument, controlled releases of ethanol (evaporated cooking wine) and acetic acid (white vinegar) were conducted at five different measurement sites. The released substances were chosen primarily based on safety and ease of handling and were not necessarily the strongest and the most spectrally distinct absorbing species within the spectral coverage of the instrument (for reader’s reference, we have provided examples of spectral signatures of different chemicals accessible by this QCL-DCS system in the supplementary material). Nevertheless, such field tests are extremely valuable for the development the DCS technology. To further validate the system performance, small volumes of a few other gaseous species were locally sprayed directly in the beam path for short bursts a few times per day.

The QCL spectrometer is based on the dual-comb technique,¹⁷ where two sources with nearly matched mode spacings are mixed on a high-bandwidth photodetector.^36–38 This mixing process generates a radio-frequency (rf) spectrum in which the absorption information is encoded. An overview of our spectrometer is shown in Fig. 1(a), and a typical deployment scenario is shown in Fig. 1(b). Figure 1(c) shows the internals of the spectrometer. It is built around two QCL frequency combs,²² which emit a few hundred milliwatts of power at ∼10 μm (1000 cm⁻¹). Both lasers along with two thermo-electrically cooled photodetectors (VIGO Systems SA) are housed in a watertight (IP66) enclosure (from now on referred to as the DCS core) to prevent contamination during operation. The lasers are actively cooled to allow for continuous operation under warm ambient conditions. The DCS core has a size of 31 × 60 × 11 cm³ (L × W × H) and a volume of ∼20 L and is equipped with two optical ports that are used as optical input and output ports [see Figs. 1(a) and 1(c)]. In addition to the lasers and detectors, the DCS core also houses beam combining and focusing optics. Part of the enclosure is reserved for future implementation of an active laser frequency stabilization scheme. One of the lasers acts as a local oscillator (LO) whose beam is confined within the DCS core for heterodyne mixing with the signal laser (SIG) on the reference photodetector marked REF in Fig. 1(a). The SIG laser is delivered through the optical output port and coupled to the telescope transceiver system where it is launched for remote sensing. Examples of retroreflector positioning in urban environments are shown in Figs. 1(b) and 1(d). The return light collected by the telescope is delivered back to the DCS core through the optical input port. This asymmetric balanced DCS configuration was implemented to take advantage of the heterodyne gain provided by the strong LO comb, which in a retroreflector-based remote sensing configuration is expected to provide enhancement to the signal and, thus, enables longer optical pathlengths. All optics inside the DCS core are cage-mounted to prevent misalignment during transportation, and after initial in-laboratory alignment, the enclosure was kept sealed throughout the deployment.

The transceiver optics used to launch and collect the light comprises beam expansion and compression optics, a motorized gimbal to aim at targets, and an imaging system for target localization. Figure 2(a) shows a schematic of the transceiver design with an inset of an IR photo of the beam taken at 50 m distance. To adhere to laser safety protocols, the laser beam launched into free space is expanded to 5 cm diameter before launch using a combination of lenses and a 3 in. 90° off-axis parabolic mirror [see Fig. 1(a)]. This ensures that the irradiance is below the eye safety limit with good margin. After expansion, the collimated beam is launched at the retroreflector target using a motorized gimbal equipped with a 6 in. flat protected gold mirror. The gimbal is controlled by software that takes input from a custom-made real-time video imaging system (see the supplementary material). This allows for user-guided target location based on a live video-feed, which drastically shortens the time needed to locate the targets. Once the targets have been identified, their positions are stored in memory, and switching between different targets takes only a few seconds (limited primarily by the gimbal speed). The retro-reflected light is collected by the same gimbal mirror and after reflection from a 50/50 beamsplitter is guided through beam compression and re-collimation optics before sent to the optical input port of the DCS core.

Upon entering the DCS core, the laser light that has interacted with the urban atmosphere (SIG) is heterodyned with the strong local oscillator (LO) on a high-bandwidth (1.75 GHz) photodetector [marked SIG in Fig. 1(a)], whose electrical output signal is digitized using a dual-channel, 12-bit, 3.2 GS/s, field-programmable gate array (FPGA) system (NI PXIe-5775). The second FPGA channel is used to digitize the REF detector signal. Both SIG and REF detectors used in the system have similar characteristics (responsivity, bandwidth, etc.) to ensure well-matched DCS signals in both channels. At the time of the deployment, the FPGA was used solely to acquire data and write it to a separate solid-state drive for post-processing analysis. Due to the high bit-rate and slow disk-writing process, the acquisitions at the time of deployment were limited to 100 µs at ∼2 Hz rate, which corresponds to an effective duty-cycle of $∼2×10−4$⁠. Due to the limited acquisition rate imposed by the computational hardware, atmospheric turbulence and rapid plume dynamics were not studied, but QCL-DCS is ideally suited for the study of fast transient phenomena.³⁸ A second detector (with matching characteristics) is used as a reference detector on which parts of the SIG and LO beams are directly mixed. The foot-print of the optical breadboard measures 61 × 92 cm,² and the electronics is housed in a separate 19 in. rack. The total system peak-power consumption is $<$700 W.

The signal post-processing analysis is divided into two steps: (1) calculation of absorbance spectra through a coherent averaging algorithm³⁹ and (2) path-integrated concentration retrieval through weighted least square (WLS) fitting (see Supplement 1) using a spectroscopic database model.^40,41

The optical spectra of the two lasers are shown in Fig. 2(b), and multiheterodyne rf spectra at two different retroreflector distances are displayed in Fig. 2(c). The green spectrum corresponds to a retroreflector distance of 14 m, and the purple spectrum corresponds to a distance of 230 m. Figure 2(d) shows the decrease in average beat note power as a function of retroreflector distance. The observed decrease for distances $>$50 m is attributed to gradually overfilling the retroreflector due to beam divergence. Although the average beat note power is reduced at longer distances, the number of detected beat notes is preserved. In practice, this means that an increase in averaging time is required to maintain the signal-to-noise ratio (SNR) at longer target distances.

The field deployment was conducted in the downtown Boston area over the course of 5 consecutive days. The sensor was operated during daytime with one measurement site per day. Presented here is a representative dataset from day 5. A map of the measurement site on Northern Ave. in Boston is shown in Fig. 3, where the sensor and the retroreflector locations are marked. A wind rose plot is shown in the upper right corner with the wind directions and speeds recorded during the release. The location of the release at each site was selected to minimize interference with pedestrians and traffic while maintaining a high likelihood of intersecting the release plume with the sensor beam. The releases were carried out by periodically evaporating ethanol (cooking-wine) and/or acetic acid (white vinegar) using portable turkey fryers [see Fig. 3]. The release rates were controlled by adjusting the turkey fryer burner and were estimated as the total volume loss per hour (ranging up to 1 gallon/hour or 3.8 l/h for the highest settings). The measurements on the day 5 of the deployment presented in Fig. 5 were obtained with the moderate release rates of <0.25 gallon/h (or ∼1 l/h). This safely mimics a realistic chemical release scenario in an urban environment with the downside of uncertainty and variability in the plume concentrations.

As part of the startup procedure of the system, an automated frequency calibration step is conducted to ensure that the wavelengths of the lasers are known. This is accomplished by switching to the frequency calibration path [see Fig. 1(a)] via a mirror placed on a motorized translation stage. In this configuration, part of the signal beam passes a hermetically sealed reference absorption cell containing an absorber with a known concentration and pressure. Bromomethane (CH₃Br) was selected as frequency calibration gas due to its periodic absorption features in the wavelength range of the lasers. A WLS fit, using the Pacific Northwest National Laboratory (PNNL) database⁴¹ with the laser offset frequency and mode spacing as free-parameters, is used to determine the frequency axis for the subsequent measurements. The weights are determined based on the reference detector signal to suppress the influence of weaker beat notes [see the lower panel in Fig. 4(a)]. Figure 4(a) shows an example of a spectrum together with a baseline corrected WLS fit. Through this procedure, the laser frequencies are periodically verified without manual intervention.

To validate the system performance, a confidence check procedure was performed a few times per day. This entailed spraying short bursts of pure chemicals stored in atomizer bottles or pressurized cans directly into the beam path. The selection of methanol (MeOH), Freon 134a (R134a), and ethanol (EtOH) as confidence check gases is based on public safety concerns, ease of handling, and availability. In addition to the aforementioned gases, acetic acid (AA) was included in the fitting procedure due to its presence in the release. The data collected during a confidence check is shown in Fig. 4(b). Figures 4(c)–4(e) show individual fits to measured spectra corresponding to the time-stamped data in Fig. 4(b).

An example of the results from a release of ethanol (cooking wine) and acetic acid (white vinegar) is shown in Fig. 5(a). The path-integrated concentrations of acetic acid, Freon 134a, methanol, and ethanol are extracted from WLS fits to the acquired spectra using the frequency axis from Fig. 4(a). Large fluctuations in the measured path-integrated concentrations are expected as the beam intersects the gas plume close to its origin before it has been adequately mixed with the surrounding atmosphere. Atmospheric turbulence therefore directly influences the beam-plume overlap, leading to large variations in measured column densities. This is clearly observed for ethanol, which exhibits peak column densities in the % m-range. The absence of peaks for acetic acid is attributed to the relatively low acetic acid content in white vinegar together with the lower volatility of this species. The bottom panel shows the signal detector power recorded during the measurements. Black and gray traces represent unfiltered and filtered data, respectively. Here, a simple threshold filter (red dashed trace) was used to discard measurements with low signal amplitude, caused by, e.g., beam blocking due to crossing traffic. To evaluate the system performance, background data at the same site was collected after the releases were stopped. The results from the fit to the background are shown to the right in Fig. 5(a). These results are used for evaluation of the system stability.

The system precision and integration time limit are estimated using Allan–Werle analysis⁴³ of the background data shown to the right in Fig. 5(a). The results are shown in Fig. 5(b), where a drift is observed after ∼100 s (200 measurements). This type of drift is often encountered in laser sensors due to optical fringe noise (etalons) whose influence can be limited by reducing thermal expansion effects, limiting the number of optical surfaces and optimizing the placement of refractive and transmissive optics. Where possible, the optical components were cage-mounted to prevent inter-component drifts due to thermal expansion, but further optimizations are possible. The 1σ noise equivalent column densities (NECDs) [ppm m] can be estimated from Fig. 5(b), which gives one second NECDs of 173 ppm m for ethanol, 59 ppm m for AA, 25 ppm m for R134a, and 14 ppm m for methanol. At 100 s of averaging, 22 ppm m for ethanol, 7.6 ppm m for AA, 4.4 ppm m for R134a, and 1.4 ppm m for methanol are obtained. The differences in NECD directly reflect the absorption cross sections of these molecules at this wavelength, and a factor of up to 10× improvement could be obtained by targeting the strongest mid-infrared absorption bands for these species. A useful metric for laser-based spectroscopic systems is the noise-equivalent fractional absorption (NEA), which can be calculated according to the procedure outlined in supplementary material. This results in a NEA of 4.8 × 10⁻² Hz^−1/2, which is primarily limited by the low effective duty-cycle dictated by a short data sampling time compared to the time required for data transfer and storage. Due to the writing-to-disk overhead, only 200 µs of data is recorded for every second of system operation time. Normalization to the effective acquisition time results in a NEA of 6.9 × 10⁻⁴ Hz^−1/2, which is in line with earlier reported results measured in laboratory settings.^36,38,44,45

The ability to detect and localize small quantities of airborne chemical releases in densely populated areas is important for issuing warnings in case of unauthorized hazardous emissions, gas leaks, and accidental chemical releases. In addition, sensor systems capable of identifying and localizing sources with complex composite gas mixtures can help to deter illicit clandestine production of drugs and explosives. Laser-based sensors are well-equipped to address the challenges of urban sensing, and the recent advancements in mid-infrared dual-comb spectroscopy provide the prospect of an autonomous sensing platform with high brightness sources for remote sensing, high temporal resolution for mobile applications, and good spectral coverage for multispecies detection, all with the possibility to enclose the sensor in a robust, hermetically sealed package without moving parts.

The first field-deployment of a mid-infrared QCL-based dual-comb spectrometer presented here is a step toward this goal. The spectrometer offers the ability to continuously monitor the surroundings up to a distance of a few blocks with 1 s sensitivities down to -14 ppm m, which are currently limited by the low effective duty-cycle of the instrument and narrow spectral coverage of the lasers. Assuming white noise limited short-term performance, a sensitivity increase by a factor of $1/(2×10−4)≈70$ is theoretically possible by increasing the system duty-cycle. Ongoing work on the FPGA platform is dedicated toward this goal, and since the time of the deployment, the duty-cycle has been increased a 100-fold, resulting in a SNR increase of ∼7. Nevertheless, the performance of this first field-deployable QCL-DCS system is quite promising, even in comparison to the results obtained with well-established remote sensing systems based on mid-infrared tunable external cavity QCLs (EC-QCL). For example, Block Engineering reported a commercial EC-QCL system,⁴⁶ LaserWarn™, that combines two miniature EC-QCLs covering a large portion of the mid-IR (7.5–12.8 μm). The LaserWarn system can also access the same absorption band of Freon-134a, and a NECD of 2.1 ppm m for a 1 s averaging time can be estimated from their reported results. This is about an order of magnitude better than the real-time NECD of 25 ppm m measured for our QCL-DCS system; however, if adjusted for the actual measurement time, a NECD of 0.4 ppm m Hz^−1/2 (estimated for 100% duty cycle) looks very promising and could potentially provide performance comparable to the LaserWarn system. In terms of NEA, the best EC-QCL remote sensing systems reported by Philips et al.^6,14,47 exhibit a NEA of 5.3 × 10⁻⁵ Hz^−1/2 per sampling point or 7.5 × 10⁻⁴ Hz^−1/2 per scan (they operate in the same wavelength range but with a much larger spectral coverage of ∼240 cm⁻¹). These are very respectable NEAs for a broadband system, and the real-time performance of our QCL-DCS system is two orders of magnitude worse. Again, the main limitation is a result of the QCL-DCS effective duty-cycle of 2 × 10⁻⁴, which translates to a comparable NEA of 6.9 × 10⁻⁴ Hz^−1/2 if estimated per scan and 100% data throughput. We believe that similar to EC-QCL systems that have been perfected for more than a decade, current QCL-DCS technology with proper engineering and optimization will also approach similar performance metrics while additionally offering true integrated spectrometer designs with no moving parts.

When we compare the capabilities of QCL-DCS to other frequency-comb based remote sensing systems in the mid-IR, we need to focus primarily on the 3–5 μm region since to our best knowledge there are not many reports extending to the longer wavelengths (8–12 μm). An earlier work using a mid-IR comb centered at 3.25 μm spanning 100 nm generated via an optical parametric oscillator (OPO) in combination with a virtual-image phased array (VIPA) spectrometer for standoff methane detection⁴⁸ yielded sensitivities down to 0.2 ppm m Hz^−1/2, reported over only a short-range (26 m). Similarly, a more recent work⁵ utilized an OPO comb ranging from 3.1 to 3.5 μm with a Fourier transform spectrometer to demonstrate an open-path NECD of 6 ppm m Hz^−1/2 for methane. The DCS technique utilizing high-coherence mid-IR frequency combs generated through DFG²⁹ has been used in a series of field works by Nathan Newbury’s group at NIST,^25,30,28 and they presented very impressive results in terms of spectral coverage, spatial coverage, and sensitivity. In addition to small molecules, the NIST group has also reported measurements of broadband absorbers, such as acetone and isopropanol³⁰ or propane and ethane.²⁵ The NECDs reported in these works ranged from several ppm m obtained over minute averaging times³⁰ down to ∼0.9 ppm m Hz^−1/2 reported for ethane.²⁵ Given comparable NECDs in the single ppm m range at minute averaging time observed for the current QCL-DCS and the prospect of improved performance down to sub-ppm m Hz^−1/2 levels with improved duty-cycle, the ability to use semiconductor laser frequency combs in the long-wave IR may offer a viable alternative for future field applications.

One of the main limitations in this first proof-of-concept field deployment was the narrow spectral coverage of 40 cm⁻¹ together with the specific experimental requirement for probing safe analytes that did not possess the strongest absorption bands within the spectral coverage of the instrument. This resulted in an increased risk of cross-interference in the WLS-fitting routine. In security and safety applications, the threshold limits for a positive detection of the individual target analytes play a critical role, but for this proof-of-concept demonstration, targeting the non-optimal benign target species, we have not attempted to determine them. The potential cross-interferences will certainly play an important role in determining these limits in the future, and a detailed crosstalk analysis needs to be performed for future detection of chemical threats in urban environments. The issue of cross-interferences can be reduced by a careful selection of the operating wavelengths of the lasers together with an increase in their spectral bandwidth.

QCL-based dual-comb developments by other research groups have been successful at extending the spectral coverage of the lasers, extending the spectral coverage of the lasers, increasing the spectral resolution of the system, as well as stabilizing and calibrating the frequency axis.^49,50 Already QCL frequency combs with more than 100 cm⁻¹ coverage have been documented,⁵¹ and recently, a QCL-based dual-comb spectrometer with a spectral resolution of ∼30 MHz over a 55 cm⁻¹ wavelength range was demonstrated.⁴⁴ 

System configuration updates to increase the sensitivity are already in progress, such as incorporating a multi-pass cell to enable point sensing capabilities to the system,⁵² adding motorized control of the beam-expanding optics to allow for optimization of the return signal at various distances, and adding a third single-mode laser as a frequency reference.⁵³ 

At present, no single sensor technology can address the dynamic and complex nature of urban trace-gas sensing, but mid-infrared dual-comb spectrometers possess the traits required to hold a central role in this application space.

See the supplementary material for signal processing details, the design and modeling of the imaging system, and examples of spectral signatures accessible within the spectral coverage of the reported dual-comb instrument.

The authors acknowledge Thorlabs for providing the lasers and Physical Sciences Inc. for handling the releases and providing the wind data. MIT Lincoln Laboratory is acknowledged for coordinating the field campaign. We also thank Glenn Atkinson for his help with the spectrometer enclosure and laser mounts.

This work was funded by the Defense Advanced Research Projects Agency (DARPA) (Grant Nos. W31P4Q-16-1-0001 and HR00111920006), Thorlabs Inc., and the open acess publication of this work was supported by the Princeton University Library Open Access Fund.

The authors have no conflicts to disclose.

Jonas Westberg: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (equal); Methodology (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Chu C. Teng: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Yifeng Chen: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Jie Liu: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Software (supporting); Validation (supporting); Visualization (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Link Patrick: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Software (supporting); Validation (supporting); Visualization (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Linhan Shen: Methodology (supporting); Validation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Michael Soskind: Data curation (supporting); Methodology (supporting); Software (supporting); Validation (supporting); Visualization (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Gerard Wysocki: Conceptualization (lead); Funding acquisition (lead); Investigation (equal); Methodology (equal); Project administration (lead); Resources (lead); Supervision (lead); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.