1. Introduction

Classical mid-infrared (MIR) spectroscopy employs dispersive or Fourier-transform infrared (FTIR) spectrometers to perform qualitative and quantitative analysis. Dispersive systems (mono- or polychromators) are relatively simple and robust but have lower signal-to-noise ratios (SNR) compared to FTIR spectrometers. The latter benefit from the multiplex and throughput sensitivity advantages (also known as Fellget’s and Jacquinot’s advantages [1]), making FTIR spectrometers the gold standard in the field. However, it has already been demonstrated that the utilization of slit masks can unlock the multiplex advantage for monochromators as well [2]. With the development of digital micromirror devices (DMD) and their prominent application in compressive sensing [3], DMDs have recently been employed to serve as a programmable reflective mask in visible [4] and near-infrared (NIR) grating-based spectrometers [5] - an approach known as spectral-coding. By using the DMD to perform spectral-coding similar to conventional slit masks, the multiplex advantage can be harnessed, however, at much higher speed and fully automated. Due to the full programmability and high flexibility in combination with the high switching rates of the micromirrors, acquisition times could be decreased down to the millisecond regime for NIR spectroscopy [6]. In the course of the last decade it has become of increasing interest to use DMDs in spectral regimes outside of their original range in the visible and NIR. Several reports of UV- and MIR modified DMDs can be found in literature. For example, UV enhanced DMDs are found to be used in spectroscopy [7], while MIR enhanced DMDs are used for applications in MIR projectors [8] and beam shaping [9].

Beyond the instrumental features and measurement concepts introduced above, the sensitivity and performances of MIR spectroscopic techniques can further be enhanced by applying laser sources instead of conventional (thermal) emitters. Nowadays, different approaches for laser-based MIR spectroscopy systems have evolved to state-of-the-art techniques in the field and they cover a huge bandwidth of applications [10]. For instance, Quantum Cascade Lasers (QCL) are well-established in various applied fields: e.g., in gas analysis [11,12] and biomedical spectroscopy [13,14], and can be used for ultra-fast MIR spectroscopic ellipsometry [15,16]. Another interesting development is their usage in frequency comb spectroscopy [17]. A relatively new class of MIR laser sources are supercontinuum (SC) lasers, which promise to enrich laser-based MIR spectroscopy [18–22]. In comparison to QCLs, they feature broader spectral bandwidths [23–26] (partly in the ranges inaccessible by QCLs), thus making them comparable with thermal emitters with the difference that the emitted radiation features high brightness and has laser properties (M² factors close to 1, i.e. nearly diffraction-limited beam qualities, directionality, spatial coherence). This unique set of emission properties coupled with ultra-broadband spectral coverage is not only of great applied interest but also introduces the Jacquinot advantage to non-FTIR systems (the core of the single mode fiber acts as a slit), thus enhancing the intrinsic sensitivity of spectroscopic instruments. Therefore, MIR supercontinuum sources are perfectly suitable for various applications, such as absorption spectroscopy (enhanced SNR or extention of path lengths) [27–29], gas sensing [30–32], hyperspectral imaging and microspectroscopy [33–37] and stand-off measurements [38,39]. Moreover, interest in SC sources and their unique properties is interdisciplinary. For instance, SC-based optical coherence tomography has been recently combined with an MIR hyperspectral imaging modality [40] that is capable of correlative morphological and chemical imaging. Hence, the combination of high-sensitivity instrumentation (simple, reliable and with intrinsic SNR advantages) with modern high-performance laser sources opens up new possibilities for MIR spectroscopy. In this contribution, a system of a DMD-based spectral-coding MIR spectrometer that employs an MIR supercontinuum source and thus exhibits both sensitivity advantages (Fellget’s and Jaquinot’s) is presented and characterized; the system’s capabilities and high flexibility are demonstrated experimentally for time-monitoring of spatial gas distributions as well as gas-cell measurements of different gases.

In IR gas sensing applications [41], usually gas bands in the NIR (overtones) are monitored and either gas contents in air columns are detected [42], multi-pass measurements are made [30] or Differential Absorption LIDAR (DiAL) is used, which determines also the spatial distribution of gas concentrations [43]. This is typically done using monochromatic lasers and works well for long path-lengths ($\sim$m - km) because a big number of absorption events leads to a high signal in the NIR regime. However, in the MIR range most atmospheric gases have distinctive absorption bands, which have high magnitude and thus may lead to total absorption for long optical path lengths. Nevertheless, these bands provide an advantage if the optical path-lengths are small ($\sim$cm), as it is for example the case in the monitoring of car exhausts. There are already existing approaches to monitor exhaust gases of passing vehicles spatially resolved, based on NIR laser diodes and DiAL [44].

In this contribution we introduce a DMD-based spectral-coding MIR spectrometer in combination with a MIR supercontinuum source. By this combination we exploit both the Fellget’s advantage (DMD spectral-coding) and the Jacquinot’s advantage (supercontinuum laser source) in the MIR. This means we benefit from the advantages of FTIR spectrometers, however, at much higher flexibility, faster acquisition time and laser-enabled stand-off ability. We demonstrate the system’s performance with gas absorbance measurements in the MIR spectral range for various gases. In particular, we determine the spectral quality by gas cell measurements. Moreover, we perform spatially- and time-resolved gas measurements in an open-path configuration, simulating the on-road monitoring of the exhaust plume of a moving vehicle.

2. Materials and methods

The experimental setup is based on a spectral-coding spectrometer that utilizes a DMD to encode the spectral information. This technique exploits the spectral multiplex-advantage in the given wavelength band [6]. In the MIR regime most IR-active gases feature distinctive absorption bands (see Fig. 1 for details). In order to combine the DMD-based spectral-coding spectrometer with the MIR SC laser source, it was necessary to equip the DMD with an MIR transparent sealing window. The original window, made of borosilicate glass, is transparent only in the NIR spectral range up to 3750 cm⁻¹ followed by strong absorption. The replacement CaF$_{2}$ window is transparent up to $\sim$1000 cm⁻¹ (Fig. 1). The housing of the employed DMD (Texas Instruments, DLP7000 0.7 XGA 2xLVDS) basically consists of two components, namely a ceramic substrate and a kovar frame holding the borosilicate glass. Only the latter has to be replaced.

 

Fig. 1. a) Emission spectrum of the supercontinuum laser source (light grey background) and absorbance spectra of the investigated gases (colored lines) taken from the PNNL database [45]; all gases feature strong absorptions in the MIR regime; b) transmission profiles of the modified MIR DMD sealing glass and the original one; the original borosilicate glass fully absorbs below 3750 cm⁻¹, the MIR replacement window exhibits high transparency over the entire MIR spectral range of the SC laser, thus enabling MIR supercontinuum-based spectral-coding spectroscopy in this regime.

Download Full Size | PDF

The basic layout of the experimental setup is depicted in Fig. 2. Two different SC sources were used. The first one was a ZBLAN fiber-based source (NKT Photonics, SuperK MIR, 2.5 MHz pulse repetition rate, 6.4% average pulse-to-pulse energy fluctuation measured within 3 nm spectral band [27] with a total average power of 200 mW radiated in the MIR part (beyond wavelengths of 2.4 µm (4200 cm⁻¹), as measured using a corresponding edge-pass filter). All CO$_2$ measurements in the spectral range between 3750 cm⁻¹ and 3550 cm⁻¹ were done using this source. The other SC source (Leukos, 400 kHz pulse repetition rate, 15% pulse-to-pulse energy fluctuation within 3 nm spectral band) is based on an InF₃ fiber, with an average power in the MIR regime (above 4200 cm^-1) of 230 mW and a pulse repetition rate of 400 kHz. This source has a broader spectral coverage and radiates up to a wavelength of 4.7 µm (around 2130 cm^-1). It was used due to its wider spectral coverage to measure different gases in gas reference cells.

 

Fig. 2. Experimental arrangement: The SC laser beam generated in the measurement unit is redirected towards the sample path (a) and is scanned in one coordinate. The gas pipe is moving orthogonal to the scanning plane, simulating a passing vehicle. After interaction with the gas cloud, the SC light is reflected back into the DMD-based MIR spectral-coding dispersive spectrometer (b). The grating splits the beam into spectral components that are projected onto the DMD; the encoded spectral distribution is focused onto a single point detector.

Download Full Size | PDF

The measurement unit (see Fig. 2(b)) consists of two sub-systems that are separated by a beamsplitter (inserted after the laser output collimator): the scanning system and the DMD-based MIR spectral-coding spectrometer. The spectrometer employs a blazed grating (Thorlabs, 300 lines/mm) to separate the spectral components; a spherical mirror (Thorlabs, f=100 mm) to image the spectrum on the MIR modified DMD that performs spectral coding for further retrieval of multiplexed spectra. A plano-convex CaF$_{2}$ lens (f=60 mm) and a ZnSe microscope objective (f=6 mm) are used to focus the encoded spectral distribution on an single-pixel MCT detector (VIGO, PCI-4TE-12 1x1). The scanning part of the system is based on a galvo mirror (Sino-Galvo) and employs a retroreflector array with gold microcube reflectors (IMOS Gubela GmbH) to ensure the efficient redirection of the scanned beam into the spectrometer. The array is positioned at a distance of 30 cm from the scanner for demonstration purposes. However, the distance is easily scalable since the use of retroreflectors and highly collimated laser sources enables complete on-axis geometry. For measurements of open-path CO$_{2}$ gas-flows, the galvo mirror system scanned the SC laser beam across the retroreflector array - perpendicular to the gas-flow (Fig. 2). After the interaction, the back-reflected beam was coupled directly into the DMD-based MIR spectrometer via the beamsplitter. For gas measurements in reference cells or the gas-flow cell the galvo mirror was kept in a fixed position and the gas cell was inserted into the beam path between retroreflector and galvo mirror. Therefore, the optical path-length is twice the length of the gas-cell. For the measurements of CO₂ a gas mixing device (QCAL Messtechnik GmbH) was used to produce different concentrations of CO₂ in N₂. For the measurements of NO₂, NO, CO, CH₄ and Formaldehyde reference gas-cells of 10 cm length were used (Wavelength References, Inc.). All the gas cells contain 100 torr of the respective gas, balanced to 500 torr with N$_2$. The CO$_2$ concentration measurements were done using a single-pass gas-flow cell with 10 cm length. All gas spectra presented are absorbance spectra calculated by taking the decadic logarithm of the ratios of the incident intensity and the transmitted intensity.

The MIR DMD spectral coding spectrometer has a standard polychromator design. A DMD modulator is employed instead of a conventional array detector. Via the DMD the spectral components are modulated (encoded) using slit-mask patterns and focused onto the detector. Each slit-mask pattern consists of one line of a Hadamard matrix that is reshaped and split in two binary Hadamard patterns. Each point in the encoded signal results from the difference of the consecutive multiplexed measurements with the two binary Hadamard slit-mask patterns. Spectra are retrieved by solving the resulting system of linear equations [6]. The spectral resolution of the spectrometer was designed by means of Zemax simulations. The SC laser beam was simulated as a polychromatic point source at infinity with a Gaussian apodization of the actual beam size (valid assumption considering the collimation and single mode fiber used for SC generation). The full width at half maximum (FWHM) of a single spectral component in the DMD plane depends on the wavelength and is 94 µm at 2380 cm⁻¹ and 100 µm at 2180 cm⁻¹, according to the design. For the given total spectral bandwidth $\Delta \lambda$ of approximately 420 nm and a DMD width of $\Delta l$=14 mm, this corresponds to a theoretical optical resolution of 3 nm, assuming linear dispersion of the grating. The actual resolution of the system $\delta \lambda$ is determined by the size of DMD super-pixels, which are grouped together by elementary DMD pixels (sub-pixels). The super-pixels are utilized to encode the spectral components. In our case, 128 super-pixels were used (i.e. 128 bars per slit-mask pattern) for all reference gas-cell measurements characterizing the system capabilities. The width of the super-pixels is $\delta l$=109.4 µm (8 sub-pixels with pitch of 13.68 µm each). Thus, the spectral resolution is defined by the sampling theorem and requires at least $n$ pixels per spectral feature to be resolved (we use $n=3$ according to the common convention defined in [46]):

Therefore, the actual spectral resolution of the system is 9.8 nm (defined in wavelengths as the dispersive system is nearly linear in this domain). The operational regime of our system lies between 3800 cm⁻¹ and 2140 cm⁻¹, this corresponds to resolutions between 6.8 cm⁻¹ and 4.5 cm⁻¹ respectively defined in wavenumbers. In the open-path scanning measurements of CO₂ only 32 super-pixels were used, thus reducing the spectral resolution by a factor of 4 whilst reducing the acquisition time by the same factor. Acquisition times per single spectrum vary between 14 ms and 3.5 ms for 128 and 32 pixels (i.e. spectral bands) respectively.

The DMD was operated at a switching rate of 18.2 kHz. While the measurement speed and thus the switching rate is less relevant for static gas cell measurements, it is of high relevance for the scanning measurements. In the gas-scanning setup 50 spatial points were measured per scan and the sampled distance along the retroreflector was 10 cm. The gas pipe was fixed on a translation stage (Zaber) in order to simulate a passing gas-emitter such as a vehicle. The total measurement time during one stage scan was 4.22 s. During this time 1200 spectra (each of 32 pixels) were accumulated (24 galvo-scans). A boxcar averager (Zurich Instruments, UHFLI 600 MHz Lock-in Amplifier with UHF-BOX option) was used to demodulate the signal. DMD trigger signals (18.2 kHz) were used to synchronize data acquisition and DMD pattern switching. The synchronization of the galvo-scanner and the rest of the system was done in post-processing. This was achieved by placing an absorber (Polyethylene, PE) on the border of the scanning range. The absorption line was then used in post-processing to determine where a new scan had started. The CO$_2$ plume images were produced by calculating the pixel-wise absorbance from a background and sample measurement and then integrating over the two CO$_2$ bands between 3750 cm⁻¹ and 3550 cm⁻¹. All absorbance spectra measured from gas reference-cells and the gas-flow cell were interpolated, baseline corrected and then smoothed using a Blackman-Harris window.

3. Results

The sensitivity of the measurement unit was characterized using the SC source from NKT photonics. Concentration rows of CO₂ were measured in order to calculate the limit of detection (LOD) of the system. The gas-mixture created in the gas-mixing device was guided through the gas-flow cell and kept at a constant flow rate of 2 normal litre per minute (NLPM). Concentrations from 0.2 % (the smallest possible with the mixing device) to 3 % CO₂ in N₂ were produced and measured. The respective absorbance spectra of the CO₂ bands between 3750 cm⁻¹ and 3550 cm⁻¹ can be seen in Fig. 3(a). The spectral distortions visible at smaller concentrations can be attributed to the measurement noise. Figure 3(b) shows a linear regression fitted to predictions of a partial least squares (PLS) model. The r² value is 0.9986 and the LOD was calculated according to:

 

Fig. 3. Characterization of the system sensitivity for real-time CO₂ scans: results of the measurements of a CO₂ concentration row. The linear regression fitted to the PLS data suggests a LOD of 0.06% (0.012% m⁻¹).

Download Full Size | PDF

In order to demonstrate the capabilities of the MIR DMD spectrometer, the various gas samples in the reference cells were measured using the SC source from Leukos that radiates up to 2130 cm⁻¹, thus also covering distinctive absorption bands of CO and N$_2$O. The resulted measurements are depicted in Fig. 4. The dashed lines in each of the four sub-figures correspond to data taken from the PNNL database [45] that have been smoothed by means of a Gaussian filter ($\sigma$ = 4.8 cm⁻¹) for illustrative purposes in order to make the spectra more comparable.

 

Fig. 4. MIR DMD spectral-coding spectrometer measurements of reference gases in 10 cm long gas cells. Dashed lines indicate reference data taken from the PNNL database. Atmospheric CO$_2$ is present in the N$_2$O spectrum at $\sim$2350 cm⁻¹.

Download Full Size | PDF

The solid colored lines are the spectra measured with the MIR spectral-coding spectrometer combined with the Leukos SC laser. Figure 4(a) shows that the measured spectrum of N$_2$O also exhibits the presence of atmospheric CO$_2$ (2380 cm⁻¹–2300 cm⁻¹). This is due to small local fluctuations of the CO$_2$ concentration in ambient air between the background and sample measurement, which were detected by the system. Further deviations from the database spectra can be attributed to measurement noise in the encoded signal. This noise is introduced by the intensity fluctuations in combination with the comparably low repetition rate of the specific source, resulting in small artifacts in the retrieved spectra. Differences in acquisition times have to be considered as well. The gas spectra in Fig. 4 are averages of 40 single spectra, leading to a total measurement time of 480 ms. The database FTIR spectra are averages of 256 interferograms per spectrum, thus resulting in acquisition times of several minutes per spectrum. In addition to measurement errors, fluctuations of atmospheric water vapor lead to distortions of the spectra. In order to cover the whole spectral region of interest the system had to be adjusted (i.e. the grating and the following optics had to be rotated) between the measurements of N$_2$O and CO and the measurements of NO$_2$ and CH$_4$.

Qualitative measurements of CO$_2$ were executed using the gas-flow cell in combination with the gas-mixer. For the spectra shown in Fig. 5(a), the SC source from NKT was used and for the measurements in Fig. 5(b) the Leukos source was employed because it delivers more power in this spectral regime.

Therefore, different Box-car and detector gain settings had to be used for the measurements with the respective SC sources. The gas concentration during measurement at 3650 cm⁻¹ was 5 % at a flow-rate of 2 NLPM. In the range of 2350 cm⁻¹ a concentration of 1 % at a flow-rate of 2 NLPM was used to prevent total absorption. The small band visible at 2200 cm⁻¹ in Fig. 5(b) is an artifact introduced by the pulse-to-pulse energy fluctuations in combination with the low pulse repetition rate of the SC source. Due to the low measurement time these factors lead to noise in the encoded signal that caused errors in the retrieved spectra.

 

Fig. 5. Measurement of CO$_2$ in the 10 cm long gas-flow cell in combination with the gas-mixer. Two distinctive CO$_2$ bands around 3650 cm⁻¹ (a) and 2350 cm⁻¹ (b) are shown. Dashed lines indicate reference data taken from the PNNL data base.

Download Full Size | PDF

Simulations of passing CO$_2$ emitting vehicles were modeled in a laboratory setting; the time-spatial evolution of the gas cloud was retrieved. A gas pipe was moved perpendicular to the retroreflector array at a distance of around 28 cm from the galvo-scanner. Each measurement consisted of 24 scans. Several runs with different CO$_2$ flow rates were made. The images of gas plumes that are shown in Fig. 6 are absorbance images. Background (reference) measurements were acquired shortly before each sample measurement. The total measurement time for each full measurement was 4.22 s. During this time the gas pipe moved 8 cm in vertical direction. The gas-flow ran perpendicular to the 10 cm line scanned by the SC laser (Y coordinate). In Fig. 6(a) a 0.5 NLPM flow was produced and the plume was measured. It is clearly visible that an amount of CO$_2$ above the detection threshold can only be seen in the first second, when the pipe was still close to the retroreflector and the CO$_2$-cloud was not yet dispersed in space. This suggests, that at such low flow-rates the CO$_2$ quickly mixes into the surrounding air. In Fig. 6(b) a higher CO$_2$ flow of 2 NLPM was used for the measurement. In this scenario higher than the surrounding CO$_2$ concentrations can be clearly detected during the whole scanning time. In Fig. 6(c) twice the flow was dialed in on the gas mixing device and the resulting image shows a much more directional CO$_2$ stream.

 

Fig. 6. Open-path CO₂ absorbance measurements of a modeled passing gas-emitter such as a vehicle; flows of 0.5, 2 and 4 NLPM have been measured with the MIR DMD spectral-coding spectrometer: a) For 0.5 NLPM the CO$_2$ concentration drops rapidly below the detection threshold due to dissipation; b) at 2 NLPM a pronounced turbulent CO$_2$ flow could be imaged, the inset depicts original unprocessed spectral data (orange solid line) from the cloud; c) 4 NLPM, a well-formed and more direct stream of higher concentration is detected and visualized.

Download Full Size | PDF

4. Conclusions and outlook

In this contribution, we introduced a MIR DMD-based spectral-coding spectrometer combined with a supercontinuum laser source. The developed system combines both multiplex (Fellget) and throughput (Jaqcuinot) advantages, typically known from FTIR instruments. While the former is introduced by the spectral-coding approach the latter is unlocked by the properties of the ultra-broadband supercontinuum emission (high beam quality, directionality, and spectral brightness). Moreover, the system features a flexible and robust design for gas sensing scenarios easily scalable for remote detection in field conditions.

The system has been characterized regarding its spectroscopic performance (limit of detection, spectral resolution and speed were accessed) and its gas sensing capabilities. Various atmospheric gases, including CO, N$_2$O, CO$_2$, CH$_4$ and NO$_2$ have been measured for that purpose and compared to reference spectra. Furthermore, a scanning approach was employed to access the temporal and spatial evolution of CO₂ gas clouds in open-path measurement mode. In order to model a passing vehicle under lab conditions, CO₂ was emitted from a moving gas pipe that was connected to a gas-mixer. The resulting spatially resolved 2-dimensional CO$_2$ integrated absorbance images demonstrate the ability of the system to perform spatially resolved monitoring of dynamic gas distributions.

The developed measurement unit combining the scanning and spectrometer sub-systems could be of relevance for various spectroscopic applications, e.g. roadside exhaust monitoring or other applications of exhaust and emission monitoring in industry and environmental science where studies of the temporal behaviour is of interest. The performance of DMD-based methods can be further advanced by targeted adaptation of the amplitude spatial-light modulation technology for the MIR spectral region. The ongoing evolution of supercontinuum laser sources, with their potential to cover the entire MIR region, is of considerable importance for further progress in this area.